See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/390221426

Machine Learning-Based Cryptocurrency Prediction: Enhancing Market

Forecasting with Advanced Predictive Models

Article in Journal of Ecohumanism · March 2025

DOI: 10.62754/joe.v4i2.6663

CITATIONS READS

0 41

7 authors, including:

Siddikur Rahman Md Abdullah Al Montaser

University of the Cumberlands University of North Texas
12 PUBLICATIONS 84 CITATIONS 7 PUBLICATIONS 30 CITATIONS

SEE PROFILE SEE PROFILE

Joy Chakra Bortty Araf Nishan

Westcliff University Mercy University
9 PUBLICATIONS 75 CITATIONS 8 PUBLICATIONS 30 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Muhammad Rafiuddin Haque on 30 March 2025.

The user has requested enhancement of the downloaded file.

 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663

Machine Learning-Based Cryptocurrency Prediction: Enhancing Market

Forecasting with Advanced Predictive Models
Md Shahidul Islam1, Monjira Bashir2, Siddikur Rahman3, Md Abdullah Al Montaser4, Joy Chakra
Bortty5, Araf Nishan6, Muhammad Rafiuddin Haque7

Abstract
The cryptocurrency market, with its record volatility and breakneck speed, is a revolutionary phenomenon that is reshaping the entire
world's landscape. Unlike regular markets, cryptocurrencies undergo unprecedented volatility caused by a complex interaction of factors
ranging from speculative trading to updates in regulations, technological innovations, and macroeconomic trends. The central objective of
this research was to develop and evaluate machine learning-driven models of cryptocurrency price trend forecasting. The focus of this
research project revolved around prominent cryptocurrencies, i.e., Bitcoin (BTC), Ethereum (ETH), and other prominent altcoins,
within the United States. The dataset employed in this analysis comprises vast historical price data, trading volumes, and key market
indicators of major cryptocurrencies, i.e., Bitcoin (BTC), Ethereum (ETH), and other major altcoins. Historical price data is presented
in terms of daily, hourly, and minute-level opening, closing, high, and low prices, providing detailed insights into temporal price behavior.
Trading volumes, which reflect the intensity of trading action, are also provided to represent liquidity and investor participation behavior.
The dataset also includes various market indicators, i.e., moving averages, relative strength index (RSI), Bollinger Bands, and other
technical indicators, which play a pivotal role in establishing market patterns and momentum. Three models are chosen in this study:
Logistic Regression, Random Forest Classifier, and XG Boost Classifier. For classification models, accuracy, precision, recall, and
F1-score metrics are employed to evaluate the performance of the models in terms of predicting the directions of the markets (e.g., upward
or downward directions). With the highest accuracy, Logistic Regression was the best-performing of the models tested, showing its relative
superiority. The integration of AI forecasts into cryptocurrency trading has the potential to revolutionize the United States financial
markets by providing traders and institutional investors with advanced tools to make decisions. The use of AI tools in cryptocurrency
trading also has significant implications for United States regulation compliance. The integration of machine learning tools within
cryptocurrency trading platforms is a significant step towards unleashing the true potential of AI in the financial markets. The field of
AI-based cryptocurrency forecasting offers numerous areas of future research with the potential to break through present limitations and
unlock new paths of market analysis. One of those areas is the use of deep learning models, i.e., Long Short-Term Memory (LSTM)
networks, for time-series cryptocurrency forecasting.

Keywords: Cryptocurrency, Machine Learning, Market Forecasting, Predictive Analytics, Financial Modeling.

Introduction

According to Islam et al. (2024b), the cryptocurrency space has grown exponentially since Bitcoin's release
in 2009, developing from a niche technological sandbox to a multitrillion-dollar asset class.
Cryptocurrencies have upended traditional financial systems by offering decentralized, borderless, and
transparent alternatives to fiat currencies. This invention has attracted a wide range of participants, from
retail traders to institutional investment houses and governments, all seeking to unlock the potential of
blockchain technology and digital assets. Khan et al. (2024) argued that while the growth of the market has
been unprecedented, it has been accompanied by extreme volatility, with the values of cryptocurrencies
experiencing double-digit percentage swings over short time frames.

In retrospect, Das et al. (2025), Bitcoin, the largest traded cryptocurrency, has seen its value fluctuate
between a few cents in its early days to an all-time high of nearly $70,000 in 2021, followed by sharp
1 MBA- Business analytics, International American University, Email: [email protected], (Corresponding Author)
2 Master of Business Administration, International American University.
3 MBA in Management Information Systems, International American University. Los Angeles, California, USA.
4 MS in Business Analytics, University of North Texas.

5 Department of Computer Science, Westcliff University, Irvine, California, USA

6 MBA in Business Analytics, International American University. Los Angeles, California, USA
7 MS in Business Analytics, Mercy University, New York, USA

2498
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
corrections. Such volatility is brought on by a combination of factors ranging from sentiment to
announcements of regulation, technological updates, to macroeconomic events (Byun & Shahbazi, 2021).
Unlike traditional assets such as stocks or bonds, cryptocurrencies have no intrinsic value and are highly
susceptible to speculative trading, thus rendering their movements unpredictable. Such unpredictability
necessitates robust forecasting tools that can help crypto economy participants navigate the dynamics of
the crypto economy (Sizan et al., 2023).

Problem Statement

As per Chouksey et al. (2023), traditional financial models, such as autoregressive integrated moving
averages (ARIMA) and generalized autoregressive conditional heteroskedasticity (GARCH), have been
applied to predict asset prices in conventional markets over the years. They rely on linear assumptions and
historical data to forecast future behavior. Their effectiveness in cryptocurrency markets is, however,
undermined by the nature of digital assets. Cryptocurrencies exchange in a low-regulated, decentralized
environment, and therefore, the markets' dynamics are influenced by a wide range of exogenous and non-
linear factors (Ali et al., 2024). For example, social media sentiment, whale movements (major trades by
large-scale traders), and technological updates such as hard forks or network upgrades have strong
influences on prices. Such influences cannot be modeled by conventional models, and hence, forecasts and
investment decisions suffer. Moreover, the high volumes and velocity of data generated by cryptocurrency
markets further hinder conventional analytical tools. Under this situation, complex predictive models must
be constructed that represent the dynamism and nature of crypto markets so that decision-making is done
in a timely fashion (Ganesh, 2024).

Research Objective

The central objective of this research is to develop and evaluate machine learning-driven models of
cryptocurrency price trend forecasting. With the power of ML algorithms, we aim to overcome the
limitations of traditional financial models and provide more accurate forecasts. Specifically, we focus on
supervised learning techniques, such as regression models, decision trees, and neural networks, that perform
exceptionally well with large data volumes and detect non-linear patterns. We employ a method of training
the models on historical price data, technical indicators, and external factors such as social media sentiment
and macroeconomic indicators. We also explore the use of ensemble methods, which combine multiple
models to provide stronger predictive power. Beyond building predictive models, we also test their
performance using measures of mean absolute error (MAE), root mean squared error (RMSE), and R-
squared values. We also provide insights into how the models could be applied in practice by investors,
traders, and financial analysts, and how they could be used to better manage risk and refine investment
strategies.

Scope and Relevance

The focus of this research project revolves around prominent cryptocurrencies, i.e., Bitcoin (BTC),
Ethereum (ETH), and other prominent altcoins, within the United States. The choice of cryptocurrencies
is motivated by market capitalization, liquidity, and influence on the general crypto environment. With a
focus on the United States, we aim to derive insights that are of high relevance to American institutions
and individuals, who exert significant influence on global crypto trends. The use of predictive modeling
here has far-reaching implications for different stakeholders. For the retail investor, accurate predictions of
prices can aid in risk reduction and return maximization in an extremely volatile setting. For institutional
investors, i.e., hedge funds and asset managers, predictive models can be employed to make allocation and
hedging decisions. Financial analysts can employ the models to make evidence-based recommendations to
clients, and the models can be employed by regulators to monitor market stability and potential risks. In
general, this study is part of the growing literature on cryptocurrency market dynamics and demonstrates
the revolutionary promise of machine learning in forecasting.

2499
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
Literature Review

Cryptocurrency Market Trends and Challenges

Gupta & Sethi (2024) reported that the cryptocurrency space is a dynamic and rapidly evolving
environment, defined by its volatility and sensitivity to a wide range of influences. One of the major drivers
of the volatility of this marketplace is investor sentiment, which is shaped by news events, social media, and
public perception. For instance, positive news, i.e., the adoption of Bitcoin by large businesses or favorable
regulatory announcements, creates sharp increases in price, and unfavorable news, i.e., security attacks or
regulatory measures, creates sharp declines. Beyond sentiment, macroeconomic influences also have a
significant impact on cryptocurrency prices.

According to Islam et al. (2025c), Macro influences, i.e., inflation rates, movements in interest rates, and
political events influence investor behavior, sending capital into or out of digital assets. For instance, when
the economy is uncertain, cryptocurrencies like Bitcoin come to be regarded as "digital gold" and appeal to
those seeking an inflation hedge (Jui et al., 2023). However, this is not always the case since cryptocurrencies
also tend to correlate with traditional markets when markets are in dislocation. Another significant driver
of crypto volatility is the impact of institutional investment and speculative trading. The entrance of
institutional players, i.e., hedge funds, asset managers, and public companies, has brought both solidity and
sophistication to the marketplace (Jagannath et al., 2021).

Poudel et al. (2023) found that while institutional investment has the potential to increase liquidity and
prevent price manipulation, it also has the potential to increase volatility through large trades and
algorithmic trading. On the other hand, speculative trading by retail traders, motivated by fear of missing
out (FOMO) or herd behavior, increases price swings and creates challenges to reliable forecasting. These
multidimensional effects indicate the complexity of cryptocurrency markets and the need for advanced
analytical software to manage their volatility.

Traditional Financial Models vs. Machine Learning for Forecasting

Traditional models, i.e., autoregressive integrated moving average (ARIMA) and generalized autoregressive
conditional heteroskedasticity (GARCH), have been traditionally used to predict prices in conventional
financial markets (Islam et al. 2025c). The models rely on linear assumptions and historical data to predict
future behavior and thus are suitable for assets with similarly stable dynamics. Their application to
cryptocurrency markets is, however, hindered by the nature of digital assets. Cryptocurrencies are traded in
a low-regulated environment, and thus, the dynamics of the assets rely on a multitude of exogenous and
non-linear factors (Rana et al., 2025). For example, social media sentiment, whale movements (major trades
by large-scale traders), and technological events such as hard forks or network updates significantly impact
prices. Such factors cannot be modeled by conventional models, and thus, the models do not provide
correct forecasts and optimal investment strategies (Ray et al., 2025).

Contrarily, Rahman et al. (2024) held that machine learning (ML) offers a flexible and robust approach to
forecasting cryptocurrency prices. ML models, i.e., decision trees, support vector machines (SVM), and
neural networks possess the capability to identify complex patterns and non-linear relationships in large
data. Such models can incorporate multiple data sources, i.e., historical data of the asset's price, technical
indicators, and exogenous factors such as news sentiment and macroeconomic indicators, to provide
improved forecasts. Moreover, Sumsuzoha et al. (2024) added that ML models learn to adapt to varying
patterns in the market, thus being suitable for the dynamic and volatile nature of cryptocurrency markets.
With the aid of ML, researchers and practitioners are capable of overcoming the limitations of conventional
models and constructing stronger predictive tools to predict crypto prices.

2500
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
Applications of Machine Learning in Financial Prediction

According to Ali et al. (2024), Machine learning has been a game-changer when it comes to financial
forecasting, offering new channels for exploring and forecasting asset prices. For cryptocurrency markets,
supervised learning techniques have been widely used to predict prices. Such techniques involve training an
ML model on labeled data, where input features (i.e., historical prices, trading volumes, and technical
indicators) are used to predict the target variable (i.e., future prices). Popular supervised learning techniques
used to predict crypto prices include linear regression, random forests, and long short-term memory
(LSTM) networks. LSTM networks, a type of recurrent neural network (RNN), work well with time-series
data as they can learn temporal dependencies and long-term patterns of movements in prices (Alnami et
al., 2025).

Aside from supervised learning, techniques of unsupervised learning, i.e., clustering and dimensionality
reduction, have been used to identify patterns in markets and segment cryptocurrencies based on the
behavior of their prices. Sentiment analysis is another important application of ML in financial forecasting
that involves exploring text data on news articles, social media, and other sources to gauge the mood of the
markets. Sentiment analysis offers valuable information on investor sentiment and allows forecasting of
movements in prices based on public sentiment (Amberrkhani et al., 2025). Technical indicators, i.e.,
moving averages, relative strength index (RSI), and Bollinger Bands, are also commonly used as input
features in ML models. Such indicators, computed based on past prices and volumes, provide information
on the behavior of markets, momentum, and volatility, thereby enabling improved forecasts. With the
incorporation of these diverse data sources and techniques, ML models can generate holistic and actionable
insights to be used by cryptocurrency market participants (Jagannath et al., 2021).

Research Gaps

Seabe et al. (2023) argued that despite the impressive advancement in machine learning-based
cryptocurrency forecasting, several research gaps remain to be addressed. One of the most critical
challenges is the lack of high-frequency forecasting models that can operate in real-time or near-real-time.
Cryptocurrency markets operate 24/7, with prices moving rapidly within seconds or minutes. Current
models, trained on daily or hourly data, may not be able to include these high-frequency dynamics, which
lowers the utility of the models for short-term trading strategies. The development of models with the
ability to process and examine data at a finer granular level, i.e., minute-by-minute or second-by-second, is
critical to improve forecasting accuracy and enable decision-making in real-time (Tanwar et al, 2021).

Tripathy et al. (2024) added that another critical research gap is the need for explainable AI (XAI) in
cryptocurrency forecasts. While ML models, particularly deep learning models, have demonstrated
improved predictive performance, their "black-box" nature renders their decision-making processes
difficult to explain. This lack of explainability is a significant barrier to adoption, particularly by institutional
investors and regulators who require transparent and interpretable insights on which to make decisions.
Shamshad et al. (2023) contended that Explainable AI techniques, such as Shapley additive explanations
(SHAP) and Local Interpretable Model-agnostic Explanations (LIME), can be used to address this
challenge by offering insights into the variables driving model forecasts. By increasing the interpretability
of ML models, researchers can build trust and confidence in their deployments, opening the door to
widespread adoption in the financial markets. Closing these research gaps will be critical to driving the field
of cryptocurrency forecasting and unlocking the full potential of machine learning in financial markets.

Data Collection and Exploration

Dataset Overview

The dataset employed in this analysis comprises vast historical price data, trading volumes, and key market
indicators of major cryptocurrencies, i.e., Bitcoin (BTC), Ethereum (ETH), and other major altcoins.
Historical price data is presented in terms of daily, hourly, and minute-level opening, closing, high, and low
prices, providing detailed insights into temporal price behavior. Trading volumes, which reflect the intensity
2501
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
of trading action, are also provided to represent liquidity and investor participation behavior. The dataset
also includes various market indicators, i.e., moving averages, relative strength index (RSI), Bollinger Bands,
and other technical indicators, which play a pivotal role in establishing market patterns and momentum.
The data is sourced from well-known cryptocurrency exchanges, i.e., Coinbase and Binance, and financial
APIs, i.e., Coin Gecko and Crypto Compare, to ensure accuracy and reliability. The sources employed
provide a robust platform to examine cryptocurrency market behavior and develop predictive models. The
dataset spans several years, providing scope to examine long-term patterns as well as short-term volatility,
and is structured to provide scope both to exploratory data analysis and to complex machine learning
applications.

Key Features Selection

S/No Key Features Description

01. Opening Price The opening price of a cryptocurrency at the beginning of

a specific period (for instance, daily, hourly, or
minute-level).
02. Closing Price The last price of a cryptocurrency at the end of a specified
period.
03. Highest Price The highest price that a cryptocurrency has reached during
a period.
04. Low Price The lowest level is reached by a cryptocurrency over a
specific period.
05. Trading Volume The number of units of cryptocurrency traded within a
specific time frame.
06. Moving Averages Average of a cryptocurrency's price over some time (e.g., 7-
day, 30-day), with two forms being the Simple
Moving Average (SMA) and Exponential Moving
Average (EMA).
07. Relative Strength Index (RSI) It is a momentum indicator that measures the rate of
change and speed of price movements, typically on
a scale of 0 to 100.
08. Market Sentiment Score Quantitative indicator of investor sentiment based on text
analysis of articles, social media posts, and other
sources.
09. Return on Investment (ROI) The percentage change in the price of a cryptocurrency
over a specific period.

Data Preprocessing

A set of common data preprocessing steps was applied in machine learning, particularly when working with
time-series or tabular data. It begins with importing libraries required from pandas and scikit-learn to be
applied in data manipulation, splitting, scaling, and encoding. The code then checks and converts a 'date'
column to 'date-time' format if it exists, keeping in mind that it might be named otherwise. Missing values
are addressed by checking for their presence and offering two options: deleting rows with any missing
values or filling in the blanks with forward fill or a constant, e.g., zero. Categorical features, illustrated by a
'symbol' column, are encoded into numerical values through Label Encoding. An example 'price movement'
target feature is generated by finding the 'close' price change direction. Finally, suitable features are selected,
data is divided into training and test sets, and numerical features are normalized using Standard Scaler so
that they have zero mean and unit variance, which is critical in many machine learning models. The resulting
training and test dataset shapes are then printed to ensure.

2502
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
Exploratory Data Analysis (EDA)

Exploratory Data Analysis (EDA) is a crucial initial step in the research process that involves systematically
investigating and summarizing a dataset to discover patterns, trends, outliers, and relationships among
variables. Through data visualization techniques (histograms, scatter plots, and heatmaps), statistical
summary measures (mean, median, standard deviation), and correlation analysis, EDA helps researchers
learn more about the structure and nature of data. In the context of cryptocurrency market research, EDA
plays a vital role in uncovering key insights, i.e., volatility patterns in prices, trading volumes, and impacts
of external drivers, i.e., sentiment or macroeconomic events. It also helps in identifying data quality
problems, i.e., missing values or outliers, that must be resolved before the application of advanced analytical
methods. By providing a platform for hypothesis generation and informing the selection of appropriate
modeling approaches, EDA ensures subsequent analysis is founded on an understood dataset and,
therefore, predictive models and research findings are enhanced in terms of accuracy and reliability.

Cryptocurrency Closing Price Trend

The implemented code snippet was intended to prepare time-series cryptocurrency price data for analysis
and visualization. It begins by transforming a 'date' column into a date-time format using pandas, required
for time-based calculations. It then designates the 'date' column as the Data Frame index, which is common
practice with time-series data to facilitate time-based indexing and efficient analysis. It then orders the data
by the date-time index explicitly to be in time order, a prerequisite for any meaningful time-series analysis
or modeling. The code then generates a line plot visualization of the 'close' price over time with a title, axis
labels, legend, and grid for legibility, providing a good summary of the cryptocurrency price trend.

Output:

Figure 1. Cryptocurrency Closing Price Trend

The graph illustrates the closing price action of a cryptocurrency between July 2018 and January 2022, with
significant volatility over time. The price initially remained quite stable, fluctuating between $0.25 and $0.75
until around mid-2020. It registered a sharp spike in early 2021, with the price going over $1.75, which
reflected the growing interest and investment in the cryptocurrency space then. Following the high point,
the price exhibited volatility with sharp declines and comebacks, prominently seen in mid-2021. In late
2021, the price stabilized somewhat at the $1.00 level, indicating a new level of equilibrium. Overall, the
data indicates the high volatility of the cryptocurrency and how market forces influence its worth.

Moving Averages and Bollinger Bands

The executed code calculated and plotted two technical analysis indicators, viz., the Simple Moving Average
(SMA) and the Bollinger Bands. It calculates the 50-day and 20-day SMAs of the 'close' price first and then
calculates the upper and lower Bollinger Bands as the 20-day SMA plus and minus two standard deviations
2503
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
of the 'close' over the previous 20 days. The code then plots the original 'close' price along with the
calculated 50-day SMA, 20-day SMA, and the Bollinger Bands, with the area between the upper and the
lower bands filled in to enhance visual clarity. This plot helps in identifying the trend of the price, the
potential levels of support and resistance, and the volatility of the price relative to its moving average.

Output:

Figure 2. Moving Averages and Bollinger Bands

The chart displays the 50-day and 20-day simple moving averages (SMA) and the closing price of a
cryptocurrency and its Bollinger Bands from July 2018 to January 2022. The blue line is the closing price,
which is quite volatile, particularly in the first surge of 2021 when it exceeded $1.75. The 50-day SMA
(dashed green line) is a smoother long-term view, and the 20-day SMA (dashed red line) closely follows the
trend of the price, providing a short-term perspective. The shaded area represents the Bollinger Bands,
which depict the volatility of the price; the bands increase when volatility is high, as in the surge in the price
in the first few months of 2021, and decrease when volatility is low, indicating low volatility in the price.
This analysis shows how moving averages and the Bollinger Bands assist in establishing the trend of the
market and the potential movements of the price, depicting the volatility of the cryptocurrency and the role
of the indicators in the eyes of a trader.

Correlation Heatmap of Features

The provided code created a heatmap to visualize the correlation matrix of the selected numerical features
in the Data Frame, 'open', 'high', 'low', 'close', 'Volume XRP', and 'Volume USDT'. It uses the seaborn
library's heatmap function to plot the pair-wise correlations in a color-coded manner, where annot=True
displays the correlation values on the heatmap. The cmap='coolwarm' parameter is used to specify the color
scheme to be used, where different colors represent the correlation's orientation and strength (positive or
negative). The visualization is useful in establishing the linear relations between different price and volume
indicators, which is important in feature selection and understanding the dynamics of the cryptocurrency
market.

2504
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
Output:

Figure 3. Correlation Heatmap of Features

The correlation heatmap displays the correlation between various features of a cryptocurrency, ranging
between -1 and 1, where values signify the strength and direction of the correlation. The highly positive
correlation is evident in the "open," "high," "low," and "close" features, with "close" and "open" having a
correlation coefficient of 0.99, meaning that the latter two tend to increase or decrease simultaneously. Also,
the "high" and "low" values of the price are highly correlated, with a correlation of 0.99, meaning uniform
movements in the price throughout the trading period. However, trading volume features, i.e., "Volume
XRP" and "Volume USDT," correlate less with the price features, with a low correlation of 0.35 with
"open." "Volume XRP" and "Volume USDT" correlate higher at 0.82, meaning a higher correlation
between the volumes compared to the prices. Overall, the heatmap shows the interdependence of the price
features and relative independence of the volumes' metrics in this cryptocurrency analysis.

Distribution of Daily Returns

The code in the Python program performed a volatility analysis by calculating the daily rate of change in
the 'close' price and storing it in a new 'Daily Return' column. It generated a histogram to plot the
distribution of the daily returns, with 50 bins and a transparent purple color. It further added a vertical
dashed red line on the histogram to mark the mean daily return and a legend to name it. The plot is then
enriched with a title and axis labels, which provide a good indication of the frequency of different daily
return values, which helps in determining the volatility and common movements of the asset's price.

Output:

2505
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
Figure 4. Distribution of Daily Returns

The distribution of the daily returns of a cryptocurrency is illustrated in the histogram with a sharp spike at
zero, indicating that the majority of the daily returns cluster in the area around this point. The x-axis is
divided into daily returns between -0.4 and 0.6, and the y-axis is divided by frequency. The distribution is
nearly symmetric with a slight tail on the left, indicating rare instances where the returns have been negative.
The red dashed line represents the mean return, which is nearly zero, further indicating that the daily
performance of the cryptocurrency has been stable over the period. The fact that the frequency is quite
high in the narrow area around zero also suggests low volatility, but the fact that extreme values on both
ends exist suggests that large swings do occur, just not as frequently. Overall, the distribution is revealing
of the risk and performance profile of the cryptocurrency.

Closing Price Distribution by Weekday

The computed code made a box plot to graph the distribution of the 'close' price across various days of the
week. It first makes a new 'Weekday' column in the Data Frame by extracting the name of the day from the
date-time index. It then uses Sea Born's boxplot function to graph it, with 'Weekday' on the x-axis and
'close' price on the y-axis. The palette is "cool warm" for styling with colors, and the order parameter is
used to get the weekdays in the correct time order. The plot is supplied with a title, axis labels, and rotated
x-axis labels to facilitate easier reading, and it provides a visual comparison of the central tendency,
dispersion, and potential outliers of the closing price on each weekday.

Output:

Figure 5. Closing Price Distribution by Weekday

The box plot illustrates the distribution of the cryptocurrency's closing price across different days of the
week, with different price movements between Monday and Sunday. The box is the interquartile range
(IQR) of the closing prices on each day, and the line inside it is the median price. Interestingly, Tuesdays
and Wednesdays have higher prices, with median closing prices of between $1.00 and $1.25, as compared
to other days, particularly on Saturdays and Sundays, where the median is closer to $0.75. The outliers,
denoted by single points outside the whiskers, represent the occasional large price spikes, particularly on
weekdays. This pattern could reflect higher trading volumes or sentiment in the middle of the week, and
weekends could be less favorable to higher closing prices, possibly suggesting low trading volumes or
participation in the markets. Overall, the data indicate weekday effects as being pivotal in cryptocurrency
trading and pricing.

2506
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
Rolling Volatility of Cryptocurrency

The implemented code calculated and printed the rolling volatility of the cryptocurrency's closing price. It
computes the 10-day rolling standard deviation of the 'close' price and writes it to a new 'Rolling Volatility'
column. It then makes a line plot showing how the rolling volatility changes over time, with the date on the
x-axis and the volatility on the y-axis. The plot is titled, axis-labeled, and legended, so it is clear that the red
line is the 10-day rolling volatility. Finally, a grid is added to the plot so it is easier to read, with a visual
representation of how the volatility of the price has changed over the period being examined.

Output:

Figure 6. Rolling Volatility of Cryptocurrency

The graph shows the 10-day rolling volatility of a cryptocurrency over the period from July 2018 to January
2022, depicting significant volatility swings in the market over this period. The volatility is low initially,
usually less than 0.05, indicating a stable market. Yet, there was a spectacular spike in volatility in early 2021,
where it went over 0.30, which was simultaneous with heightened trading volumes and market attention
toward cryptocurrencies. This volatility spike suggests very high price volatility and investor uncertainty,
which is likely driven by market speculation and exogenous shocks. Subsequently, the volatility comes down
but is higher than in the first episode, indicating a volatile market environment even after the initial spike.
Overall, the graph shows the cryptocurrency's susceptibility to sudden changes in market conditions,
emphasizing the importance of monitoring volatility to facilitate risk management and trading.

MACD Indicator for Trend Momentum

The computed code calculated and plotted the Moving Average Convergence Divergence (MACD)
indicator, a popular indicator of trend momentum. It begins by calculating the 12-period and 26-period
Exponential Moving Averages (EMAs) of the 'close' price. It then calculates the MACD line by subtracting
the 26-period EMA from the 12-period EMA. It then calculates a 9-period EMA of the MACD line as the
'Signal Line'. The code then plots the MACD line (blue) and the Signal Line (red and dashed) over time,
along with a horizontal dotted gray line at zero. This plot helps traders to identify potential buy and sell
signals based on the MACD and Signal line crossovers and the position of the MACD relative to the zero
line.

2507
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
Output:

Figure 7. MACD Indicator for Trend Momentum

The graph presents the MACD (Moving Average Convergence Divergence) indicator and signal line from
July 2018 through January 2022, which are critical tools for identifying trend momentum within the
cryptocurrency market. The MACD line, plotted in blue, fluctuates around the zero line, highlighting zones
of bullish and bearish momentum. Notably, a large spike occurs in early 2021, with the MACD exceeding
0.2, reflecting strong bullish momentum during this time, perhaps a reflection of the broader market hype
surrounding cryptocurrencies. The red dashed signal line provides a smoother alternative, with buy and sell
signals based on crossovers with the MACD line. Over the period under observation, several crossovers
occur, particularly in late 2020 and early 2021, reflecting potential trading opportunities. Overall volatility
of the MACD points to a dynamic market environment, underscoring the importance of the MACD as a
trend-following momentum indicator to traders looking to profit from the price movements.

Liquidity Analysis Price Spread vs Volume

The code fragment performed a basic liquidity analysis by comparing the price spread with the trading
volume. It calculated the daily spread in terms of 'high' and 'low' prices and put this in a new 'Spread'
column. It then makes a scatter plot using seaborn where 'Spread' is on the x-axis and 'Volume USDT' on
the y-axis. The plot includes blue markers with some transparency so that data points are visible. The title
and axis labels indicate that the plot is meant to explore how the trading volume is related to the price
spread, and this could help ascertain the liquidity of the market at different ranges of prices. It includes a
grid to enable easier reading of the scatter plot.

Output:

2508
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
Figure 8. Liquidity Analysis Price Spread vs Volume

The scatter plot analyzes the correlation between trading volume and price spread (as the difference
between high and low) of a cryptocurrency, with both being plotted on the x and y axes. The x-axis indicates
the price spread, ranging between 0 and slightly over 1 USD, and the y-axis indicates trading volume, up to
4 billion USD. The bulk of data points cluster in the bottom left quadrant, indicating that low price spreads
have higher trading volumes, indicating higher liquidity in those situations. With a growing price spread,
the number of data points decreases, indicating higher spreads being supported by decreasing trading
volumes. This trend indicates the inverse correlation between liquidity and price spread, where tighter
spreads tend to enable higher trading volumes. Overall, the chart highlights the importance of price spread
as a key driver of liquidity in the cryptocurrency market, which is an important consideration when assessing
markets.

Methodology

Feature Engineering

Feature engineering is a crucial step in the machine learning pipeline that transforms raw data into
meaningful inputs that enhance the predictive power of models. In predicting cryptocurrency prices, some
of the most predictive features include moving averages, Relative Strength Index (RSI), trading volume,
and sentiment scores. Moving averages, i.e., Simple Moving Average (SMA) and Exponential Moving
Average (EMA), smoothen price volatility and identify trends over a specific time horizon. These features
help models filter out the noise in the short term from the price movement in the long term. The RSI, a
momentum oscillator, provides information about overbought or oversold conditions, which are critical in
predicting potential reversals in the price. Trading volume, a measure of market activity, is an indicator of
liquidity and investors' interest and tends to be associated with significant price changes. Sentiment scores,
based on textual analysis of news articles and social media, put a number on the emotional tone of the
market, incorporating a qualitative element into the dataset.

Lagged features are produced to better represent short-term and long-term patterns to further enhance the
predictive ability of the models. Lagged features involve the use of past data at previous time steps (for
example, past 7, 30, or 90 days' worth of volumes or prices) as input to predict future values. For example,
last night's closing price or last week's average trading volume might be helpful context to describe current
movements in the price. Lagged features enable models to include temporal dependencies and cyclical
patterns in the data, particularly applicable in highly volatile markets like cryptocurrencies. With the addition
of these engineered features, the dataset is enriched and rendered more informative, enabling machine
learning models to pick up on complex patterns and provide better forecasts.

Model Selection and Training

The use of machine learning models for cryptocurrency price prediction is based on selecting models that
can capture the unique nature of the market. Three models are chosen in this study: Logistic Regression,
Random Forest Classifier, and XG Boost Classifier. Logistic Regression is included as a baseline for
predicting price directions (upward or downward trends) based on linear relationships between input
features and the target. Simple as it may be, the model provides a baseline for measuring the performance
of sophisticated models. Random Forest Classifier, as an ensemble of decision trees, is included to capture
non-linear relationships and interaction among features. Random Forest aggregates the decisions of several
trees to prevent overfitting and generalize well to new, unseen data. XG Boost Classifier, as a high-
performing gradient boosting algorithm, is included due to its high predictive performance and efficiency
in handling large data. XG Boost constructs decision trees in a sequential manner, where each tree attempts
to correct the mistakes of the preceding tree, and thus it develops a highly optimized model.

The use of ensemble learning techniques, i.e., Random Forest and XG Boost, is backed by the argument
that they can harness the strengths of multiple models and offset each other's weaknesses. Ensemble
techniques improve predictive capacity by reducing variance, boosting robustness, and modeling diverse
2509
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
patterns in data. For instance, a decision tree may overfit noise in data, but an ensemble of trees averages
out such mistakes, making more accurate predictions. Using these models, the study aims to achieve a trade-
off between explainability (Logistic Regression), flexibility (Random Forest), and performance (XG Boost),
allowing comprehensive coverage of the problem space.

Model Optimization and Performance Analysis

For maximum possible accuracy and robustness of the selected models, hyperparameter tuning is done
using Grid Search, an exhaustive method of evaluating different combinations of hyperparameters. Grid
Search systematically tries predefined ranges of hyperparameters, e.g., the number of trees in Random
Forest or the learning rate in XGBoost, to determine the optimal setup of each model. This process ensures
that the models are tuned to the nuances of the cryptocurrency dataset, enhancing their predictive capability.
Additionally, cross-validation is applied to test the models' ability to generalize and prevent overfitting.
Cross-validation divides the dataset into several folds, trains the model on part of the data, and then
validates it on the rest of the folds. This approach provides a better estimate of the performance of the
models by reducing the impact of data variability and ensuring the models generalize well to different
subsets of data.

The optimization also includes an evaluation of the model's performance on both the training and validation
data to check for overfitting. Overfitting occurs when a model learns to identify noise in the training data
rather than underlying patterns, leading to poor performance on new data. Through careful hyperparameter
tuning and the use of cross-validation, the study ensures that the models find a balance between complexity
and generality, leading to stable and reliable predictions.

Evaluation Metrics

The performance of the models is evaluated using a combination of regression metrics when predicting
prices and classification metrics when forecasting movements in the markets. For regression tasks, Mean
Squared Error (MSE) and Root Mean Squared Error (RMSE) are used to measure the accuracy of the
projected prices relative to actual prices. MSE computes the average square of the difference between actual
and projected values, and RMSE provides a better understandable measure by finding the square root of
MSE. Such measures are particularly useful when assessing the accuracy of projected prices and finding
models that minimize errors in projections.

For classification models, accuracy, precision, recall, and F1-score metrics are employed to evaluate the
performance of the models in terms of predicting the directions of the markets (e.g., upward or downward
directions). Accuracy measures the proportion of correct predictions, and precision and recall measure the
ability of the models to predict positive cases (e.g., upward directions) and not misrepresent false positives
or false negatives, respectively. The F1-score, being the harmonic mean of precision and recall, provides a
balanced measure of performance, particularly when dealing with imbalanced data where one class is
dominant. A comparison of performances is carried out to evaluate the strengths and weaknesses of both
models. Logistic Regression is interpretable and easy to understand, but it may not be capable of capturing
complex non-linear relationships in data. Random Forest, being an ensembling method, is better in terms
of accuracy and reliability but may be computationally intensive. XG-Boost has advanced optimization
techniques.

2510
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
Results and Analysis

Cryptocurrency Market Trend Analysis

Figure 9. Cryptocurrency Market Trend

The graph shows the size of the growth of the cryptocurrency market, split between hardware and software
segments, estimated between 2020 and 2030 in terms of USD billions. The stacked bar graph shows a
steady growth in the size of the total market over the decade, with hardware (dark purple segment) and
software (light blue segment) both contributing to the growth. The software segment is seen to have a
slightly greater share in the size of the market than hardware in the majority of the years estimated. The
graph also shows a global market Compound Annual Growth Rate (CAGR) of 13.1% between 2025 and
2030, with significant expected growth in the second half of the estimated period. The data is reported as
being taken from Grand View Research.

Model Performance Evaluation:

Logistic Regression Modelling

The code utilizes a Logistic Regression classification algorithm, perhaps modeling the 'price movement'
target set up in earlier steps. It initially imports scikit-learn modules, the Logistic Regression class, Grid
Search CV for hyperparameter tuning, and metrics to be used in the evaluation. It initializes a Logistic
Regression model with the 'liblinear' solver. It then creates a parameter grid to search over values of the
regularization strength ('C') and the type of penalty ('penalty': 'l1' and 'l2'). It then utilizes Grid Search CV
to execute a grid search over this parameter grid to discover the hyperparameter combination that achieves
the best performance on the training set. After it fits the grid search, it prints the best parameters and
utilizes the model with the best parameters to predict the scaled test set. It then utilizes a classification
report, a confusion matrix, and the accuracy score to assess the performance of the tuned Logistic
Regression model and prints it to measure the effectiveness of the model.

2511
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
Output:

Table 1. Logistic Regression Results

precision recall f1-score support

0 0.69 0.22 0.34 129

1 0.53 0.90 0.67 128

accuracy 0.56 257

macro avg 0.61 0.56 0.50 257
weighted avg 0.61 0.56 0.50 257

Accuracy for Logistic Regression: 0.5603

Accuracy for Logistic Regression: 0.5603

The above table shows the performance metrics of an Optimized Logistic Regression model with specific
parameters, i.e., a regularization strength (C) of 10 and penalty type 'l1'. Precision, recall, and F1-score values
of class 0 are 0.69, 0.22, and 0.34, respectively, indicating that the model is very good at classifying class 0
instances but is poor in recall, suggesting a high number of false negatives. For class 1, the low precision is
0.53, recall is 0.90, and thus the F1-score is 0.67, indicating better detection of class 1 instances. The
accuracy of the model is 56.03%, with a total support of 257 instances in both classes. Macro average
metrics suggest well-balanced detection across classes, despite the low recall of class 0, suggesting scope for
improvement. The confusion matrix indicates that the model misclassified 29 class 0 instances as class 1
and 13 class 1 instances as class 0, and thus the need to improve further in discriminating between the two
classes.

Random Forest Classifier Modelling

The implemented code employed a Random Forest Classifier for prediction, again perhaps on the 'price
movement' target variable. It employs the scikit-learn required classes, i.e., Random Forest Classifier, Grid
Search CV, and evaluation metrics. The Random Forest Classifier is initialized with a set random state to
provide reproducibility. A large hyperparameter grid is defined to tune the number of trees (n-estimators),
maximum tree depth (max_depth), the number of samples required to split an internal node
(min_samples_split), and the number of samples required to be at a leaf node (min_samples_leaf) 1 2.
GridSearchCV is then employed to perform 5-fold cross-validation on all possible combinations of the
hyperparameters to find the set that provides maximum performance on the training data. The optimal
hyperparameters learned by GridSearchCV are then printed, and the trained model with the optimal
hyperparameters is employed to predict the scaled test data. Finally, the accuracy of the tuned Random
Forest is tested and reported using a classification report, confusion matrix, and overall accuracy score.

2512
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
Output:

Table 2. Random Forest Classifier Results

precision recall f1-score support

0 0.46 0.46 0.46 129

1 0.46 0.47 0.47 128

accuracy 0.46 257

macro avg 0.46 0.46 0.46 257
weighted avg 0.46 0.46 0.46 257

Accuracy for Random Forest: 0.4630

The table above presents the performance metrics of a Random Forest model, with precision, recall, and
F1-score provided for two classes, 0 and 1. For both classes, the precision is 0.46, indicating that the model
is correct in 46% of positive calls. The recall is also 0.46 for class 0, indicating that the model is catching
just 46% of true class 0 instances, and class 1 is slightly better at 0.47 recall, indicating it has a slightly better
ability to detect true positives in that class. The F1-scores for both classes are identical at 0.46, indicating
both precision and recall are well matched, yet overall performance is low. The accuracy of the model is
46.30%, with total support of 257 instances in both classes, indicating much potential for improvement.
The confusion matrix indicates that the model is misclassifying 59 class 0 instances as class 1 and 68 class
1 instances as class 0, again indicating the challenge in being able to tell the two classes apart.

XG-Boost Classifier Modelling

The code script utilized an XG Boost classifier, yet another effective algorithm well-suited to classification
tasks like predicting 'price movement'. It utilizes the XG Boost package and the scikit-learn modules
required to select the model and evaluate it. An XG Boost classifier is initialized with a random state to
provide reproducibility, disables the label encoder, and provides the evaluation function as 'log loss'. It
initializes a parameter grid to search over the key hyperparameters such as number of trees (n-estimators),
maximum depth of each tree (max depth), learning rate (learning rate), proportion of samples used when
training each tree (subsample), and proportion of features used when constructing each tree
(colsample_bytree). GridSearchCV is then used to determine the optimal set of these hyperparameters using
5-fold cross-validation on the scaled training data. The optimal set of parameters discovered by Grid Search
CV is then printed, and the corresponding model is used to predict the scaled test data. Finally, the
performance of the model is tested and reported using a classification report, confusion matrix, and
accuracy score.

Output:

Table 3. XG-Boost Results

precision recall f1-score support

0 0.46 0.49 0.47 129

1 0.45 0.41 0.43 128

accuracy 0.45 257

macro avg 0.45 0.45 0.45 257
weighted avg 0.45 0.45 0.45 257

Accuracy for XGBoost: 0.4514

2513
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
The table above provides the performance metrics of an XG-Boost model, showing its recall, precision,
and F1 score on two classes, 0 and 1. The precision of class 0 is 0.46, indicating that 46% of class 0 instances
it classified as correct, and class 1 is slightly less at 0.45. For recall, class 0 is at 0.49, indicating the model is
correct on 49% of true class 0 instances, and class 1 is at 0.41, indicating it is correct on 41% of true
positives. The respective F1 scores are at 0.47 for class 0 and 0.43 for class 1, indicating a moderate level
of balance between recall and precision. The accuracy of the overall model is at 45.14%, with 257 instances
supported in both classes. The confusion matrix shows misclassifying 63 class 0 instances as class 1 and 75
class 1 instances as class 0, indicating huge challenges in being correct in discriminating between both
classes, indicating the need to further tune the model.

Comparison of All Models

The implemented code focused on comparing the accuracy of Logistic Regression, Random Forest, and
XG-Boost classification models. It first initializes a dictionary named model comparison to store the names
of the models and corresponding accuracy values, which were possibly calculated in the earlier code cells.
It then converts the dictionary to a pandas DataFrame named comparison_df to make it easier to handle
and visualize. The Data Frame is sorted by the 'Accuracy' column in descending order so that the best-
performing model is seen readily. Finally, a horizontal bar chart is generated using matplotlib to visually
compare the accuracies of the three models, with the accuracy values on the bars to make it easier to
compare. This provides an efficient and effective summary of the performance of the different classification
models on the problem at hand.

Output:

The bar chart is a comparison of the accuracy of three machine learning models: Logistic Regression,
Random Forest, and XG Boost. With the highest accuracy of 56.03%, Logistic Regression was the best-
performing of the models tested, showing its relative superiority. Next, the Random Forest model achieved
46.30% accuracy, which was close but reflected that it captures market trends well enough, but not as well
as Logistic Regression. XG Boost, which had the lowest ranking with 45.14% accuracy, did not live up to
expectations in its predictive capability despite being an advanced algorithm. Overall, the graph is a clear
indication of how each of these models has performed based on accuracy, and Logistic Regression proved
to be the most reliable model for the dataset analyzed.

The results of the comparison of the models shed light on the importance of a feature importance analysis
in establishing which indicators have a major impact on movements in cryptocurrency prices. Examining
the features used in the models, i.e., moving averages, RSI, trading volumes, and sentiment scores, we can
establish which factors have the biggest impact on movements in price. Knowledge of said contributions
is critical to traders and analysts who must make decisions based on behavior within markets. The trade-
offs in selecting a model, however, should also be considered, i.e., the trade-off between explainability and
2514
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
predictive power. While models like Logistic Regression provide greater transparency and explainability,
allowing stakeholders to be aware of the reasons behind the forecasts, they may not be as predictive as
intricate models like XG Boost and Random Forest. Thus, the choice relies on the balance of the strengths
of transparent insights compared to the need for robust forecasts, ultimately establishing the decision-
making process within a dynamic and volatile environment.

Practical Applications in the USA

Impact on Financial Markets

The integration of AI forecasts into cryptocurrency trading has the potential to revolutionize the United
States financial markets by providing traders and institutional investors with advanced tools to make
decisions. For crypto traders, machine learning (ML) models deliver actionable data on patterns of change
in prices, enabling them to spot profitable points of entry and exit with greater accuracy. With predictive
analytics, traders can anticipate movements in markets, i.e., bullish or bearish, and adjust strategies
accordingly. This is particularly helpful in the highly volatile cryptocurrency market, where steep
movements in prices over short durations can accrue massive profits or losses. Institutional players, i.e.,
asset managers and hedge funds, also stand to benefit from AI-driven forecasts by incorporating them into
portfolio management and risk evaluation. For example, ML models are capable of picking up patterns of
correlation between cryptocurrencies and traditional assets, enabling investors to diversify portfolios and
mitigate risks. Predictive analytics also enhances risk management strategies by providing warnings of
impending market downturns or liquidity crises. With machine learning insights being woven into
processes, financial institutions can make data-driven decisions that drive higher profitability and reduce
exposure to market uncertainties.

Besides, AI tools can be employed to create algorithmic trading strategies where trades are executed
automatically based on pre-decided criteria. Such strategies could be based on predictive models and real-
time data to capitalize on arbitrage opportunities and inefficiencies in the markets. For instance, ML-driven
algorithmic strategies could analyze order book data, trading volumes, and price movements to execute
high-frequency trades with low latency. This not only makes trading efficient, but also assists in providing
liquidity and price discovery in the markets. Overall, the application of AI-driven forecasts in the United
States' markets has the potential to make the trading environment better informed, efficient, and robust,
both at the level of individual traders and institutional players.

The Regulatory Impact of Trading in the USA

The use of AI tools in cryptocurrency trading also has significant implications for United States regulation
compliance. Cryptocurrencies come under the purview of oversight agencies such as the Securities and
Exchange Commission (SEC) and the Commodity Futures Trading Commission (CFTC), which implement
regulations to protect investors and maintain market integrity. AI predictive models have the potential to
assist market participants in complying with regulations by providing transparent and auditable insights into
trading activities. For example, ML models can be used to identify transactions that demonstrate signs of
market manipulation, e.g., wash trading or spoofing, and label the activities as suspect so that further
investigation is feasible. This is in line with the SEC's mission to prevent fraud and ensure fair and orderly
markets.

Apart from fraud detection, AI tools can also assist in maintaining market stability by identifying system-
wide threats and providing warnings of potential disruptions. For instance, predictive models can scan
markets to identify signs of excessive speculation or a lack of liquidity and enable regulators to undertake
preventive measures to stabilize the market. This is particularly important in the cryptocurrency markets,
where the lack of central oversight and speculative trading pose potential volatility. With the use of AI in
regulatory compliance and market monitoring, U.S. regulators will be better equipped to regulate quickly
evolving cryptocurrency markets and protect investors from potential threats. Secondly, the design of
explainable AI (XAI) techniques will be able to mitigate concerns over the explainability and transparency
of machine learning models. Explainable AI explains why the models make certain predictions, so it is easier
2515
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
for regulators to assess the reliability and accuracy of AI tools. This is critical in building trust in AI tools
and having them used responsibly within the markets. With cooperation between tech providers, market
participants, and regulators, the United States is capable of creating a regulatory environment that fosters
innovation without compromising market integrity.

Integration with Trading Platforms

The integration of machine learning tools within cryptocurrency trading platforms is a significant step
towards unleashing the true potential of AI in the financial markets. One of the most promising applications
is the use of ML-based trading bots that automatically execute trades based on predictive models and pre-
programmed strategies. The bots can scan large data volumes in real-time, ranging from trading volumes
and price movements to market sentiment, to identify profitable opportunities and make trades with
minimal intervention. For example, a trading bot could use an ML model to predict short-term price
movements and make buy or sell orders accordingly, maximizing returns and minimizing risk. Automated
trading bots find special utility in the cryptocurrency markets, where prices fluctuate rapidly and trading is
done 24/7. Apart from trading robots, predictive analytics is also used in automated investment strategies,
i.e., robo-advisors, which suggest customized investment recommendations based on an individual's risk
profile and investment goals.

With the incorporation of ML models, robo-advisors can suggest improved and dynamic recommendations,
dynamically updating portfolios based on changing market scenarios. This brings complex investment
strategies within everyone's reach, enabling retail investors to utilize the same complex tools used by
institutional investors. Besides, predictive analytics can be used to fortify risk management on
cryptocurrency exchanges by identifying potential threats and vulnerabilities. For instance, ML models can
analyze trading behavior to identify signs of market manipulation or liquidity risk so that exchanges can
take proactive measures to protect users. Predictive analytics can be used to optimize margin requirements
and collateral management, reducing the risk of defaults and keeping the trading platform stable. With
machine learning in their business, cryptocurrency exchanges can create a safer and better trading
environment, encouraging trust and confidence in users.

Discussion and Future Directions in the USA

Challenges in AI-Based Cryptocurrency Forecasting

Notwithstanding the significant advancement in AI-driven cryptocurrency forecasting, several challenges

persist that hinder the development and application of predictive models. One of the major challenges is
the rampant existence of manipulative activities and false volumes of trading in the cryptocurrency markets.
Unlike the highly regulated traditional financial markets, the cryptocurrency platform is a decentralized
network that is not uniformly regulated, making it susceptible to manipulative tactics such as wash trading,
spoofing, and pump-and-dump. Such activities distort trading volumes and price action, injecting noise into
data and affecting the accuracy of predictive models. For instance, false volumes of trading mislead models
to recognize false trends, hence providing incorrect forecasts. Overcoming this challenge necessitates the
design of sophisticated anomaly detection tools that identify and filter out manipulative activities so that
models learn on clean and genuine data.

One of the biggest challenges is the volatility and extreme price movements that characterize cryptocurrency
markets. While volatility offers profit opportunities, it also renders it unpredictable, as the prices change
rapidly over short intervals of time due to speculative trading, announcements, or macroeconomic events.
Traditional machine learning models, which learn from historical data, may not be capable of dealing with
sudden and unprecedented movements in prices and thus suffer in terms of predictive accuracy. To combat
this issue, researchers must explore adaptive models that adapt dynamically to reflect changing market
conditions in real-time. Also, incorporating external data sources, i.e., news sentiment and macroeconomic
data, can provide context to movements in price and make predictive models stronger. The challenges
indicate the necessity of constant innovation and optimization of AI-driven forecasting techniques so that
they remain effective in the highly dynamic cryptocurrency environment.
2516
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
Limitations of the Study

While this study provides valuable information on leveraging machine learning to predict cryptocurrency
prices, it is not without limitations. One limitation is data availability on less well-known or smaller
cryptocurrencies, referred to as altcoins. While well-known cryptocurrencies such as Bitcoin and Ethereum
have a rich history and high liquidity, smaller cryptocurrencies have low volumes and short histories. This
limited data prevents the effective creation of predictive models since machine learning models require
large and diverse data to learn meaningful patterns. Additionally, data on smaller cryptocurrencies could be
of low quality with gaps or inconsistencies affecting the performance of models. Overcoming this limitation
requires higher-quality and fuller data, as well as techniques to train models on sparse or incomplete data.

Another limitation is the potential biases in predictive models caused by unpredictable events in the
markets. Cryptocurrency markets rely on a wide range of factors, from regulatory announcements to
technological innovations and political events, many of which cannot be foreseen or quantifiable. For
example, an unanticipated regulatory crackdown or a major security breach initiates large-scale movements
in the price that diverge from historical patterns, leading to incorrect projections. Such unpredictable events
introduce uncertainty into the process of forecasting, and it is hard to develop models that are always
reliable. To overcome this limitation, future research could include incorporating event detection systems
in real time that are capable of detecting and responding to unanticipated market events, making predictive
models more dynamic.

Future Research Directions

The field of AI-based cryptocurrency forecasting offers numerous areas of future research with the
potential to break through present limitations and unlock new paths of market analysis. One of those areas
is the use of deep learning models, i.e., Long Short-Term Memory (LSTM) networks, to time-series
cryptocurrency forecasting. LSTMs are a type of recurrent neural network (RNN) that is well-suited to
modeling temporal dependencies and long-term patterns in serial data and thus ideal to apply on
cryptocurrency price movements. With the assistance of LSTMs, researchers can construct models that
better capture the complex and non-linear dynamics of cryptocurrency markets, with improved predictive
power. One of the areas of potential is incorporating social media sentiment analysis into predictive models.
Social media outlets, such as Twitter and Reddit, have a significant impact on the sentiment of the
cryptocurrency markets, with opinion leaders and groups influencing the action in the markets through
opinion and discussion. With sentiment analysis, predictive models will be capable of sensing the emotional
sentiment of the markets and identifying patterns that may not be visible through price data. For example,
an increase in positive sentiment on social media could be predictive of an imminent rally in the price, and
conversely.

Advanced NLP techniques, such as transformer models (e.g., BERT), can be used on social media data to
extract actionable insights to be applied in forecasting markets. Aside from that, tremendous potential lies
in creating AI-driven investment advisory systems tailored to the requirements of financial institutions.
Such systems can be capable of leveraging machine learning models to provide investment
recommendations, make portfolio allocations, and manage risk in real-time. For instance, an AI-driven
advisory system could assess a client's risk appetite, investment objective, and current market scenario to
recommend a diversified basket of cryptocurrencies and traditional assets. Such systems, through
automation, bring efficiency, reduce expenses, and provide improved results to investors. Additionally, the
integration of explainable AI (XAI) techniques holds the potential to introduce transparency to the
decision-making process, building confidence and trust in users.

Conclusion

The central objective of this research was to develop and evaluate machine learning-driven models of
cryptocurrency price trend forecasting. The focus of this research project revolved around prominent
cryptocurrencies, i.e., Bitcoin (BTC), Ethereum (ETH), and other prominent altcoins, within the United
States. The dataset employed in this analysis comprises vast historical price data, trading volumes, and key
2517
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
market indicators of major cryptocurrencies, i.e., Bitcoin (BTC), Ethereum (ETH), and other major altcoins.
Historical price data is presented in terms of daily, hourly, and minute-level opening, closing, high, and low
prices, providing detailed insights into temporal price behavior. Trading volumes, which reflect the intensity
of trading action, are also provided to represent liquidity and investor participation behavior. The dataset
also includes various market indicators, i.e., moving averages, relative strength index (RSI), Bollinger Bands,
and other technical indicators, which play a pivotal role in establishing market patterns and momentum.
Three models are chosen in this study: Logistic Regression, Random Forest Classifier, and XG Boost
Classifier. For classification models, accuracy, precision, recall, and F1-score metrics are employed to
evaluate the performance of the models in terms of predicting the directions of the markets (e.g., upward
or downward directions). With the highest accuracy, Logistic Regression was the best-performing of the
models tested, showing its relative superiority. The integration of AI forecasts into cryptocurrency trading
has the potential to revolutionize the United States financial markets by providing traders and institutional
investors with advanced tools to make decisions. The use of AI tools in cryptocurrency trading also has
significant implications for United States regulation compliance. The integration of machine learning tools
within cryptocurrency trading platforms is a significant step towards unleashing the true potential of AI in
the financial markets. The field of AI-based cryptocurrency forecasting offers numerous areas of future
research with the potential to break through present limitations and unlock new paths of market analysis.
One of those areas is the use of deep learning models, i.e., Long Short-Term Memory (LSTM) networks,
to time-series cryptocurrency forecasting.

References
Ahmad, T., & Abbas, M. (2024). AI Based Cryptocurrency Price Prediction: A Comparative Analysis of Traditional & Deep
Learning Models with Sentiment Integration (Master's thesis, UIS).
Akula, R. & Pinky, J. N., (2024). Enhancing Cryptocurrency Market Forecasting: Advanced Machine Learning Techniques
and Industrial Engineering Contributions. arXiv preprint arXiv:2410.14475.
Ali, J., Alhussein, M., Hassan, S., Aurangzeb, K., & Umair, M. (2024). An improved machine learning-driven framework for
cryptocurrencies price prediction with sentimental cautioning. IEEE Access.
Alnami, H., Mohzary, M., Assiri, B., & Zangoti, H. (2025). An Integrated Framework for Cryptocurrency Price Forecasting
and Anomaly Detection Using Machine Learning. Applied Sciences, 15(4), 1864.
Amberkhani, A., Bolisetty, H., Narasimhaiah, R., Jilani, G., Baheri, B., Muhajab, H., ... & Shubbar, S. (2025, March).
Revolutionizing Cryptocurrency Price Prediction: Advanced Insights from Machine Learning, Deep Learning and
Hybrid Models. In Future of Information and Communication Conference (pp. 274-286). Cham: Springer Nature
Switzerland.
Byun, Y. C. & Shahbazi, Z., (2021). Improving the cryptocurrency price prediction performance based on reinforcement
learning. IEEE Access, 9, 162651-162659.
Chouksey, A., Shovon, M. S. S., Tannier, N. R., Bhowmik, P. K., Hossain, M., Rahman, M. S., ... & Hossain, M. S. (2023).
Machine Learning-Based Risk Prediction Model for Loan Applications: Enhancing Decision-Making and Default
Prevention. Journal of Business and Management Studies, 5(6), 160-176.
Das, B. C., Sarker, B., Saha, A., Bishnu, K. K., Sartaz, M. S., Hasanuzzaman, M., ... & Khan, M. M. (2025). Detecting
Cryptocurrency Scams in the USA: A Machine Learning-Based Analysis of Scam Patterns and Behaviors. Journal
of Ecohumanism, 4(2), 2091-2111.
El Mahjouby, M., Bennani, M. T., Lamrini, M., El Far, M., Bossoufi, B., & Alghamdi, T. A. (2024). Predicting market
performance using machine and deep learning techniques. IEEE Access.
Ganesh, D., & Harshavardhan, V. C. (2024, May). Cryptocurrency Market Dynamics: A Machine Learning-Based Approach
for Price Prediction. In 2024 International Conference on Smart Systems for applications in Electrical Sciences
(ICSSES) (pp. 1-6). IEEE.
Gupta, S. P., & Sethi, M. (2024, December). Predictive Modeling in Cryptocurrency: A case Study of Bitcoin using RNN. In
2024 1st International Conference on Advances in Computing, Communication and Networking (ICAC2N) (pp.
635-641). IEEE.
Islam, M. R., Nasiruddin, M., Karmakar, M., Akter, R., Khan, M. T., Sayeed, A. A., & Amin, A. (2024). Leveraging Advanced
Machine Learning Algorithms for Enhanced Cyberattack Detection on US Business Networks. Journal of Business
and Management Studies, 6(5), 213-224.
Islam, M. Z., Islam, M. S., Al Montaser, M. A., Rasel, M. A. B., Bhowmik, P. K., & Dalim, H. M. (2024). EVALUATING
THE EFFECTIVENESS OF MACHINE LEARNING ALGORITHMS IN PREDICTING
CRYPTOCURRENCY PRICES UNDER MARKET VOLATILITY: A STUDY BASED ON THE USA
FINANCIAL MARKET. The American Journal of Management and Economics Innovations, 6(12), 15-38.
Islam, M. Z. ., Islam, M. S., das, B. C. ., Reza, S. A. ., Bhowmik, P. K. ., Bishnu, K. K. ., Rahman, M. S. ., Chowdhury, R., &
Pant, L. . (2025). Machine Learning-Based Detection and Analysis of Suspicious Activities in Bitcoin Wallet
Transactions in the USA. Journal of Ecohumanism, 4(1), 3714 –. https://doi.org/10.62754/joe.v4i1.6214
Jagannath, N., Barbulescu, T., Sallam, K. M., Elgendi, I., Okon, A. A., McGrath, B., ... & Munasinghe, K. (2021). A self-
adaptive deep learning-based algorithm for predictive analysis of bitcoin price. IEEE Access, 9, 34054-34066.

2518
 Journal of Ecohumanism
2025
Volume: 4, No: 2, pp. 2498 – 2519
ISSN: 2752-6798 (Print) | ISSN 2752-6801 (Online)
https://ecohumanism.co.uk/joe/ecohumanism
DOI: https://doi.org/10.62754/joe.v4i2.6663
Khan, M. T., Akter, R., Dalim, H. M., Sayeed, A. A., Anonna, F. R., Mohaimin, M. R., & Karmakar, M. (2024). Predictive
Modeling of US Stock Market and Commodities: Impact of Economic Indicators and Geopolitical Events Using
Machine. Journal of Economics, Finance and Accounting Studies, 6(6), 17-33.
Jui, A. H., Alam, S., Nasiruddin, M., Ahmed, A., Mohaimin, M. R., Rahman, M. K., ... & Akter, R. (2023). Understanding
Negative Equity Trends in US Housing Markets: A Machine Learning Approach to Predictive Analysis. Journal
of Economics, Finance and Accounting Studies, 5(6), 99-120.
Mohaimin, M. R., Das, B. C., Akter, R., Anonna, F. R., Hasanuzzaman, M., Chowdhury, B. R., & Alam, S. (2025). Predictive
Analytics for Telecom Customer Churn: Enhancing Retention Strategies in the US Market. Journal of Computer
Science and Technology Studies, 7(1), 30-45.
Poudel, S., Paudyal, R., Cankaya, B., Sterlingsdottir, N., Murphy, M., Pandey, S., ... & Poudel, K. (2023). Cryptocurrency
price and volatility predictions with machine learning. Journal of Marketing Analytics, 11(4), 642-660.
Seabe, P. L., Moutsinga, C. R. B., & Pindza, E. (2023). Forecasting cryptocurrency prices using LSTM, GRU, and bi-
directional LSTM: a deep learning approach. Fractal and Fractional, 7(2), 203.
Tanwar, S., Patel, N. P., Patel, S. N., Patel, J. R., Sharma, G., & Davidson, I. E. (2021). Deep learning-based cryptocurrency
price prediction scheme with inter-dependent relations. IEEE Access, 9, 138633-138646.
Tripathy, N., Hota, S., Mishra, D., Nayak, S. K., Dalai, S. S., & Tripathy, J. (2024, November). Cryptocurrency Price
Prediction: A Comprehensive Analysis of High-dimensional Features with Deep Learning Techniques. In 2024
International Conference on Intelligent Computing and Sustainable Innovations in Technology (IC-SIT) (pp. 1-
6). IEEE.
Rana, M. S., Chouksey, A., Hossain, S., Sumsuzoha, M., Bhowmik, P. K., Hossain, M., ... & Zeeshan, M. A. F. (2025). AI-
Driven Predictive Modeling for Banking Customer Churn: Insights for the US Financial Sector. Journal of
Ecohumanism, 4(1), 3478-3497.
Ray, R. K., Sumsuzoha, M., Faisal, M. H., Chowdhury, S. S., Rahman, Z., Hossain, E., ... & Rahman, M. S. (2025). Harnessing
Machine Learning and AI to Analyze the Impact of Digital Finance on Urban Economic Resilience in the USA.
Journal of Ecohumanism, 4(2), 1417-1442.
Rahman, A., Debnath, P., Ahmed, A., Dalim, H. M., Karmakar, M., Sumon, M. F. I., & Khan, M. A. (2024). Machine learning
and network analysis for financial crime detection: Mapping and identifying illicit transaction patterns in global
black money transactions. Gulf Journal of Advance Business Research, 2(6), 250-272.
Sabeena, J., & Sagar, P. L. (2023, November). Enhancing Predictive Accuracy for Real Time Cryptocurrency Market Prices
with Machine Learning Techniques. In 2023 International Conference on Sustainable Communication Networks
and Application (ICSCNA) (pp. 1288-1292). IEEE.
Shamshad, H., Ullah, F., Ullah, A., Kebande, V. R., Ullah, S., & Al-Dhaqm, A. (2023). Forecasting and Trading of the Stable
Cryptocurrencies With Machine Learning and Deep Learning Algorithms for Market Conditions. IEEE Access,
11, 122205-122220.
Sizan, M. M. H., Chouksey, A., Miah, M. N. I., Pant, L., Ridoy, M. H., Sayeed, A. A., & Khan, M. T. (2025). Bankruptcy
Prediction for US Businesses: Leveraging Machine Learning for Financial Stability. Journal of Business and
Management Studies, 7(1), 01-14.
Sizan, M. M. H., Das, B. C., Shawon, R. E. R., Rana, M. S., Al Montaser, M. A., Chouksey, A., & Pant, L. (2023). AI-Enhanced
Stock Market Prediction: Evaluating Machine Learning Models for Financial Forecasting in the USA. Journal of
Business and Management Studies, 5(4), 152-166.
Sizan, M. M. H. ., Chouksey, A. ., Tannier, N. R. ., Jobaer, M. A. A., Akter, J. ., Roy, A. ., Ridoy, M. H. ., Sartaz, M. S. ., &
Islam, D. A. (2025). Advanced Machine Learning Approaches for Credit Card Fraud Detection in the USA: A
Comprehensive Analysis. Journal of Ecohumanism, 4(2), 883 –. https://doi.org/10.62754/joe.v4i2.6377
Sumsuzoha, M., Rana, M. S., Islam, M. S., Rahman, M. K., Karmakar, M., Hossain, M. S., & Shawon, R. E. R. (2024).
LEVERAGING MACHINE LEARNING FOR RESOURCE OPTIMIZATION IN USA DATA CENTERS: A
FOCUS ON INCOMPLETE DATA AND BUSINESS DEVELOPMENT. The American Journal of Engineering
and Technology, 6(12), 119-140.
Sun, G. (2024). Cryptocurrency price prediction based on Xgboost, LightGBM and BNN. Applied and Computational
Engineering, 49, 273-279.
Zhao, F., Zhang, M., Zhou, S., & Lou, Q. (2024). Application of Deep Reinforcement Learning for Cryptocurrency Market
Trend Forecasting and Risk Management. Journal of Industrial Engineering and Applied Science, 2(5), 48-55.

2519

View publication stats