Stock Market Prediction Using Machine Learning

Algorithms
Ifeoluwa Rahmon Eze Samson
Information Technology Information Technology
University of Bouremouth University of Bouremouth
Bournemouth, United Kingdom Bournemouth, United Kingdom
[email protected] [email protected]

Abstract— The stock market is the central element of movements is very appealing due to its capacity to analyze
modern capitalism, requiring shrewd decision-making in the extensive datasets, detect nuanced patterns, and adapt to
face of its unpredictability. Machine learning (ML) has become dynamic market conditions [5], [6]. The convergence of
a potential technique for addressing the complexities of stock finance and ML presents both innovative investment
market prediction, surpassing conventional approaches such as
fundamental and technical analysis. This literature review
opportunities and intriguing challenges for scholars in the
examines the ML approaches used for predicting stock market field.
movements. This study examines existing approaches, datasets, The aim of this study is to provide a comprehensive
and ML algorithms in this field, with the goal of providing a literature review on the use of ML techniques in predicting
helpful resource for researchers, financial professionals, and stock market trends. This study examines the many
investors who want to improve their knowledge of financial methodologies, datasets, and ML techniques used within
markets. The paper examines the limits of traditional this particular domain. The objective of our endeavor is to
methodologies and explores many ML methods, including provide a valuable repository for scholars, experts in the
Neural Networks, Support Vector Machines (SVM), Naive sector, and investors who possess an inclination towards
Bayes (NB), Convolutional Neural Networks (CNN), and Deep
Learning (DL). It demonstrates the effectiveness and
using ML techniques to enhance their understanding and
drawbacks of these methods in forecasting stock market proficiency in comprehending and maneuvering the
patterns. This thorough examination reveals deficiencies in intricacies of financial markets. This will be achieved by a
current research, emphasising the need for stronger and more systematic analysis and amalgamation of the extant
practical solutions, thereby directing future research efforts scholarly works.
towards more reliable prediction models.

Keywords—Stock Market, Machine Learning, Prediction, II. LITERATURE REVIEW

Artificial Neural Network
Stock market prediction (SMP), a subject that has been
widely studied, has traditionally depended on conventional
I. INTRODUCTION approaches, namely fundamental and technical analysis [7].
The stock market is a multifaceted and intricate financial Originating from the work of Fraser [8], fundamental
framework that has been often referred to be the "epicenter analysis focuses on assessing a company's true worth by
of contemporary capitalism." The primary aim of investors carefully scrutinising its financial statements, the calibre of
and traders worldwide is to capitalize on the ever-changing its management, and the trends within its industry.
market conditions. Efficient decision-making in the stock Technical analysis, as proposed by Vesela [9], utilises
market necessitates the use of sophisticated resources and historical market data and chart patterns to predict short-
methodologies, owing to the inherent volatility and term price fluctuations by identifying trends and market
unpredictability of this financial domain. The field of psychology. However, the introduction of ML methods has
machine learning (ML), which falls under the umbrella of caused a significant change in the field, leading to a rise in
artificial intelligence, is shown significant potential in alternative approaches for predicting stock market trends.
addressing these challenges [1], [2]. Throughout history, the Recent research has utilised regression models [10], time-
prediction of stock market performance has garnered series analysis using ARIMA and long short-term memory
significant attention from scholars, economists, and those networks (LSTM) [11]–[13], classification models like
involved in financial investments. Extensive study has been decision trees and support vector machine (SVM) [14], and
conducted on the predicting of stock prices, market trends, DL methods involving neural networks and CNNs [15].
and financial indicators, resulting in a wide range of These approaches have shown encouraging outcomes,
methodologies being explored. Over the years, investors exhibiting improved prediction ability. Nevertheless, upon
have always depended on fundamental and technical closer examination, there are ongoing research gaps and
research, alongside macroeconomic information, to guide concerns that continue despite these developments. It is
their decision-making processes [3], [4]. Nevertheless, these crucial to thoroughly examine these gaps and unsolved
methodologies possess some limitations, particularly in the difficulties in the current literature in order to guide future
context of handling extensive and diverse datasets, adapting research efforts towards more reliable and practical
to market fluctuations, and dealing with unorganized and solutions in predicting stock market trends. Hence, this
noisy data. Due to these limitations, there has been an section of this review paper intends to critically examine
increase in the inclination towards using ML techniques for existing literatures on both traditional and ML approach of
the purpose of predicting stock market trends. The use of SMP.
ML models for the purpose of forecasting stock market

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE

A. Traditional Approaches to Stock Market Prediction The primary objective of fundamental analysis is to
1) Fundamental Analysis ascertain the inherent worth of a company, so assessing
An essential part of stock market research, fundamental whether it is undervalued, overpriced, or appropriately
analysis looks at a number of different aspects of a priced in the market [17]. Through evaluating both
company's core operations to determine its true worth [16]. qualitative and quantitative criteria, investors get a thorough
This approach takes a holistic look at a wide range of comprehension of a company's genuine value, enabling
quantitative and qualitative aspects: them to make well-informed choices that match with their
Company Financials: A basic analysis is a comprehensive long-term investment goals. Investors use fundamental
examination of a corporation's financial statements, which research to identify firms with robust fundamentals and
includes scrutinising its income statement, balance sheet, growth prospects, with the goal of constructing portfolios
and statement of cash flows. An extensive understanding of that can withstand short-term market volatility [18].
a company's financial well-being and effectiveness may be Although fundamental analysis offers useful insights, it does
gained by examining its sales, expenditure, debt ratios, and not include market attitudes or short-term market
profitability statistics. movements. Its effectiveness depends on precise data
Management and Corporate Governance: Assessing the interpretation and accurate projections [19].
proficiency and authenticity of the management team is 2) Technical Analysis
essential. Gaining insight into the company's long-term Technical analysis is based on the premise that past market
prospects and sustainability may be achieved by data, namely price fluctuations and trade volumes, exhibit
comprehending its strategic choices, leadership style, and discernible patterns that may be used to predict future price
corporate governance practises. trends [20]. Considering existing literature on technical
Industry Trends and Market Position: Placing a firm in indicator analysis of stock market, a number of articles
the context of its industrial landscape allows for an disregarded the index's components and other big size firms
assessment of its competitive position. Market share, in favour of using alternative market indexes. In their S&P
industry growth, technical improvements, and regulatory 500 forecasting model, Lohrmann et al. [21] used this
dynamics are key factors that provide valuable insights into strategy, along with fundamental technical parameters and
prospective possibilities and hazards. returns from six main markets for stocks (DAX, Nikkei 225,
Economic Indicators: It is essential to take into account HSI, FTSE 100, Euro STOXX 50, Russell 2000), two
wider economic variables such as interest rates, inflation, exchange-traded funds (ETFs)—Vanguard Total World
GDP growth, and geopolitical developments. The success of Stock and iShares MSCI Emerging Markets—and the
a firm may be greatly affected by these macroeconomic volatility index (VIX). Additionally, the S&P 500 may be
variables, which have the power to influence its income, forecasted by Rosillo et al. [22] using three VIX-based
costs, and market demand. factors. Using the return data of ten main stock market
indices, including the projected ones,

Figure 1.0: Flowchart of the traditional approach analysis

Hoseinzade and Haratizadeh [23] forecast five of these overbought/oversold circumstances, and probable trend
indices. In order to forecast the daily movement of the reversals.
IBOVESPA index, the Bayesian Network of Malagrino et Volume Analysis: Trading volume, which represents the
al. [24] incorporates a comparable number of key stock quantity of shares exchanged within a certain timeframe,
market indexes. They used a collection of twelve different complements price research. Volume patterns often indicate
stock markets to make their forecasts. In their 2011 the robustness or fragility of price fluctuations, supporting
publication, Na and Sohn [25] forecast the KOSPI by trend analysis. The main objective of technical analysis is to
analysing the trend data of eight international stock market develop short-term trading techniques and accurately time
indexes: the FTSE 100, the DJIA, the SSE, the HSI, and the the market, enabling traders to determine the optimal
DAX. The projections made by Huang et al. [26], on the moments to enter and leave positions. It focuses on the
other hand, are limited to only one stock market. This group psychology of individuals participating in the market and
of writers specifically incorporates the S&P 500 index into how their combined actions influence changes in prices.
their methodology for Nikkei 225 forecasting. Tsai and Although technical analysis offers useful insights into price
Hsiao [27] do the exact opposite; they forecast equities movements and market emotion, it is not without
exchanged on the Taiwan Stock Exchange using data from restrictions. The use of historical data presupposes the
the TAIEX. recurrence of market behaviour, a premise that may not
Hence, this approach uses a diverse range of tools and always be valid. In addition, technical analysis may fail to
strategies to analyse and make sense of market data: take into account fundamental variables that influence the
Charts and Patterns: Technical analysts use many types of value of a company, so possibly disregarding important
price charts, including line charts, graphs of bars, and candle market-moving events or changes in investor mood.
charts, to visually represent past price fluctuations as shown Therefore, it is often used in combination with other
in figure 2.0. Recognising repetitive patterns like as head analytical techniques to create a holistic investing plan.
and shoulders, triangles, double tops/bottoms, and
support/resistance levels assists in forecasting probable
future price fluctuations.

Figure 2.0: Stock market charts, patter and candlesticks

Trend Analysis: Evaluating trends is essential in technical

analysis. Trends, classified as upward trends, downward
trends, or horizontal movements, provide valuable
information about the future direction of price movements.
Trendlines and moving averages are used to ascertain and
confirm the orientations of trends.
Indicators and Oscillators: Technical analysts use many
indicators such as the Relative Strength Index (RSI), and
Stochastic Oscillator to assess market momentum,
 Table 1.0: Previous works on Traditional Approach
Author Used Indicators Data Time frame Period Market Index
Source
Luo et al. [28] SMA, MV, MVR, Stock Daily January 2007–July SSE
EMA, 10 2008
Disparity,

Agrawal et al. [29] RSI, SMA, MFI, Stock Daily 2007–2008 NASDAQ
MACD

Boyacioglu and Avci Republic gold selling Stock Daily January 1990– ISE
[30] price, December
exchange rates 2008
Treasury bill, etc.
Esfahanipour and MA and BIAS Stock 6 days July 18, 2003– TSE and TEPIX
Aghamiri [31] December
31, 2005

Kara et al. [32] SMA, RSI, MACD and Stock Daily January 1997– ISE
Momentum December
31, 2007.

Atsalakis et al. [33] SOP, SCP and SMA Stock Daily April 2007– NBG
November 2008,

Shen et al. [34] Not mentioned Stock Next-day January 2000– NASDAQ,
October 2012 S&P500 and
DJIA

Wei et al. [35] SMA, RSI volume Stock Daily 2000–2005 TAIEX
Ratio,
psychology line (PSY),
V

Anthony et al. [36] U.S. treasury bonds Stock Daily U.S. treasury
and bonds and
treasury bills treasury bills

Olaniyi et al. [37] SMA Stock Monthly January 2007 and NSE
June 2008

Guresen et al. [38] Not mentioned Stock Daily October 7, 2008– NASDAQ
June 26,
2009

Shom and Padhy [39] EMA and difference in Stock Not January 2007– INSE
the mentioned December
percentage of the price 2010
(RDP)

Enke et al. [40] Money Supply, T-bill- Stock Daily January 1, 1980– S&P 500
rate, Proof of January
Deposit rate 1, 2009

Sureshkumar and SCP, high price, SOP, Stock Intraday October 2007– NSE
Elango [41] and low October 2011
Price
Thanh et al. Vietnamese stock market Stock Daily January 2010–April EXP, and CPI
[42] index 2016
Chong et al. Not mentioned Stock Daily January 2010– Korea KOSPI
[43] December
2014
Zhou et al. EMA, SOP, RSI, SCP, Stock Not mentioned January–December China Stock Market
[44] SMA, MACD 2016
Chen and Hao Ratio (VR), SMA, MTM, Stock Intraday [5, 10, October 31st 2008– SSE & SZSE
[45] EMA, 15, 20, Dec 31st COMP SUB
RSI, OBV and 30] days 2014 IND
MACD Volume

Thanh et al. Vietnamese stock market Stock Daily January 2010–April EXP, and CPI
[42] index 2016

B. Machine Learning (ML) Approach for Stock Market

Prediction
a study conducted by Wang [46], the author explored the
This section provides a comparative analysis of challenges associated with using a substantial amount of
several methods used for forecasting the trajectory of stock data and the unpredictability of discrepancies between
the stock market. The two primary categories of SMP two continuous time series while making predictions about
tools consist of prediction-based approaches and the stock market. In order to address these problems, the
clustering-based techniques. Prediction-based author devised a system that reduced the storage needs of
approaches encompass various techniques, including the stock price data in terms of size (e.g., MB) and using
artificial neural networks (ANN), convolutional neural fuzzification techniques in conjunction with Grey theory.
networks (CNN), and naive bayes (NB). While The approach they developed is known as the fuzzy Grey
clustering-based approaches include several system, which has the ability to rapidly forecast the stock
techniques, including filtering, k-means, and fuzzy. price at a certain moment. In their study, Lai et al. [47] used
an innovative approach that integrated k-means clustering, a
1) Analysis based on clustering techniques genetic algorithm (GA), and a fuzzy decision tree (FDT) to
In this part, the analysis is conducted using ring approaches predict the stock price of the Taiwan Stock Exchange. The
that have been established in the existing literature. The GAFDT framework first used k-means clustering to derive
description of the study is provided as follows. sub-clusters of stocks inside the Taiwan Stock Exchange
Corporation (TSEC). Subsequently, the GA was utilised to
a) Fuzzy based prediction techniques identify the most optimum fuzzy word for each input index
In the realm of fuzzy logic, the variables' truth values are in the FDT. The FDT was used to generate decision rules,
not limited to conventional binary values, but rather include with the prediction being carried out as the last step. In their
a broader range of positive and negative real integers. This study, Anbalagan and Maheswari [48] examined the
paper provides an overview of many research that use efficacy of a fuzzy meta graph (FM) in predicting stock
fuzzy-based methodologies for stock market forecasting. market trends and addressing challenges related to the non-
Esfahanipour and Aghamiri [31] created a Neuro-Fuzzy stationarity and non-linearity of time-series data. Efendi et
Inference System with the objective of predicting stock al. [49] contended that current SMP models remain
prices. The technical indices were inputted into the fuzzy inadequate in addressing the inherent unpredictability,
model, resulting in an output that was a linear combination volatility, and uncertainty associated with stock prices
of the inputs. The ruleset was separated by the use of a during data processing. Using this as inspiration, they
customized variant of fuzzy C-Means (FCM) clustering. The created an innovative forecasting technique known as the
membership function was represented by a Gaussian fuzzy random auto-regression (FR-AR) model. Within the
function in the model. ANFIS tries to predict the future FR-AR, the low-high and stock price data are managed,
trajectory of stock prices by considering many elements while the auto-regressive component addresses the issue of
such as political influences, macroeconomic shifts, technical data stationarity.
indicators, and fundamental study. The primary objective of
the first phase of Enke, et al.'s [40] three-stage approach for b) Filtering based prediction techniques
predicting stock market trends is to use multiple regression The process of filtering has generated significant attention
analysis in order to identify the relevant financial factors. In and has the potential to be used in many predictive
scenarios. This section provides an overview of many
studies that have used filtering techniques for the purpose of a) Support Vector Machine (SVM)
predicting stock market trends. The establishment of trading Given the numerous challenges associated with ANNs,
regulations was undertaken by Arévalo, et al. [50], who including the presence of locally optimal solutions, over-
derived them from the recognition of flag patterns, including fitting, and time complexity, as pointed out by various
significant insights from previous research endeavors. A authors [53]–[55], it is not surprising that researchers in the
dynamic window system is first developed to regularly field of stock market forecasting have also explored
execute the stop loss and compute the profit. In addition, the alternative prediction models. SVMs are extensively used
flag pattern adheres to prevailing market trends, prompting for classification, while SVR is often utilised for regression.
the inclusion of an indicator known as Exponential Moving As SVMs are grounded on the idea of structural risk
Average (EMA) to enhance trade filtering capabilities. The minimization, they have the ability to decrease the
calculation of technical indicators is conducted on two generalisation error with pattern of margin, support vectors,
different time periods, namely 15-minute and 1-day and hyperplane as shown in figure 3.0 [56], [57].
intervals. This approach enables the incorporation of both Consequently, several studies have been published on
short-term and medium-term viewpoints. While the model financial market forecasting using SVM- and SVR-based
did provide specific trading rules, its effectiveness in models. These publications will be further elaborated on in
mitigating data snooping was limited. Ariyo, et al. [51] the following paragraphs.
developed an ARIMA model for the specific objective of Lu [58] introduced an innovative method for predicting
stock market forecasting. stock index values by integrating nonlinear independent
component analysis (NLICA), SVR, and particle swarm
c) K-means based clustering technique optimisation (PSO). In this study, the NLICA technique was
K-means clustering is a popular vector quantization used to extract the pertinent features related to non-linearity
method used in the field of data mining for doing cluster characteristics. Additionally, the PSO algorithm was utilised
analysis. This section describes the research study that to optimise the parameters in the SVR model. Rustam and
used the K-means based stock market forecast system: Kintandani [59] conducted a study where they used a SVR
Data mining was used by Nanda, et al. [52] to create with PSO to forecast the values of Indonesian stocks.
groups of equities with similar characteristics. Following Nevertheless, the hybrid approach developed by Das and
completion of the categorization, stocks were selected Padhy [55], which integrates SVM with teaching-learning-
from the groupings to be included in the portfolio. based optimisation (TLBO), demonstrated superior
Diversifying one's holdings meets the need for lower risk. performance compared to the PSO+SVR model in
The stock market is segmented into groups according to forecasting the closing price of the COMDEX commodities
clustering criteria. The BSE dataset's valuation ratios are futures index.
employed with the stock returns. Portfolio managers and Yeh et al. [54] proposed a novel methodology called
investors may now use the clustering approach to choose MKSVR, which combines SVR with a multiple-kernel
companies that will help them reach effective frontier. (MK) learning algorithm. This method addresses the issue
of manually adjusting the hyper-parameters of the kernel
2) Analysis based on prediction techniques functions in SVR. In 2019, Gowthul and Baulkani [60]
This section presents an analysis of prediction suggested an amalgamated MK-SVM technique. Factor
methodologies like SVM, ANN, NB, and CNN with analysis was used in this model to exclude unnecessary
more reviews on previous works done with them in characteristics from the stock index data, while fruit fly
predicting stock markets as mentioned in table 2.0. optimisation (FFO) was utilised to fine-tune the parameters.

Figure 3.0: SVM framework structure

The suggested technique exhibited favourable outcomes performance and its determining variables considering its
when compared to the MKSVM, and GA+MKSVM models layers as shown in figure 4.0 [69]. Stock returns may be
in forecasting the daily price movements of equities from predicted using a number of different sets of input factors,
the NYSE, DJIA, and S&P 500. as discussed in the literature. Before feeding data into an
The authors Yu et al. [61] introduced a forecasting model ANN for prediction, some scientists transformed the raw
called the evolving least squares support vector machine input variables. This section describes the several studies
(LSSVM) learning paradigm with a mixed kernel. This that used ANNs to anticipate stock market trends, including:
model was designed to anticipate stock market movements. For the purpose of forecasting the stock market's behavior,
This approach included the use of GA on two occasions. Ticknor, [63] developed a model called Bayesian
Firstly, GA was employed to identify the crucial regularized ANN. Each stock's future value is estimated
characteristics from the dataset. Secondly, GA was utilised using current market conditions and technical indicators.
to optimise the parameters of the LSSVM. The developing The best way to help investors maximize their stock profits
LSSVM model has superior efficiency and robustness when is to provide them with accurate price predictions. Here, the
compared to other LSSVM models. This is particularly network's ability to handle complex models is given some
evident when using various individual kernels, such as consideration by the Bayesian regularized network.
polynomial, radial basis kernel (RBF), sigmoid kernel, and
mixed kernel. In addition, Chai et al. [56] suggested using c) NB based prediction technique
the LSSVM technique together with empirical mode Classification using the Nave Bayes algorithm generates
decomposition (EMD) for predicting the CSI 300 index. The Bayesian Networks for a given dataset using Bayes theorem.
EMD-LSSVM approach included testing four-parameter A feature in a class is assumed to be independent of all other
optimisation techniques, namely simplex, grid search (GS), features in the provided dataset. The NB algorithm beats
PSO, and GA. The findings demonstrated that the GS-based more complicated classification methods while also being
EMD-LSSVM method beat other methods in accurately simple to implement. The following is a discussion of the
predicting stock price changes. study done using NB for predicting the stock market. In
The research conducted by Kao et al. [62] had the objective addition, the Neuro-linguistic programming (NLP)
of identifying an appropriate wavelet sub-series for a technique was utilized by More et al. [64] to collect data on
predictive model that would enhance the accuracy of stocks and generate stock charts that investors may use to
predictions. They introduced a new prediction model named zero in on the most lucrative opportunities. Accessing
Wavelet-MARS-SVR. This approach used a wavelet massive amounts of stock data in parallel using the Hadoop
transform to breakdown the financial time-series data, architecture, and then using the NB algorithm to make
multivariate adaptive regression splines (MARS) to identify choices in the face of probability interference and the
relevant variables derived from the wavelet transformation, knowledge for anticipating future trends, was possible.
and SVR to conduct the prediction.

b) ANN-based prediction techniques

When compared to other statistical approaches, ANN is
superior at capturing the structural link between stock

Figure 4.0: ANN framework structure [63]

 d) Convolutional Neural Network (CNN) based
prediction techniques

The CNN is an example of a feed-forward neural

network. A traditional neural network does not have as
many hidden layers as a CNN does. The famous deep
learning method used for stock market forecasting is
CNN. This section addresses the studies that use CNN to
anticipate the stock market, and they are as follows:
Employing the Standard & Poor's 500 index and
technical signals, Vargas, et al. [65] used a DL technique
to make directional predictions. The complex patterns in
the data are identified and analyzed using DL
technologies, which speeds up the trading process. The
approach did not modify reinforcement learning
algorithms for use in market simulation training. By
customizing the LSTM and CNN for training to
anticipate regular stock market patterns, Zhou, et al. [66]
created a general framework.
e) Deep Learning (DL)
In their study, Cao and Wang [67] showcased the
capability of a CNN to handle both categorical and
continuous variables in the context of financial
forecasting. They successfully achieved accurate
prediction outcomes with this approach. The
researchers built two models, a CNN and a CNN-
SVM, to forecast stock index values. The empirical
findings demonstrated the effectiveness of both
models. In addition, Wen et al. [68] proposed a new
method for reconstructing time-series data using
common patterns, namely motifs. They then used a
CNN model to identify and learn the fundamental
patterns, ultimately reconstructing the time-series
sequences. This strategy was effective in predicting the
direction of stock prices. The network was trained
using the stochastic optimisation algorithm (Adam) in
this suggested technique. The LSTM network is a
sophisticated DL technique. Fischer and Krauss [69]
conducted the first investigation on the efficacy of the
LSTM model in predicting stock market trends, as
documented in the literature. Chung and Shin [70]
later combined it with a GA to enhance the LSTM
network's performance in predicting financial market
trends. This integration aimed to optimise the LSTM
architecture and establish the appropriate time window
size. In a study conducted by Baek and Kim [71], they
focused on addressing the issue of over-fitting. They
proposed a new forecasting method that utilises a
LSTM network. The ModAugNet consists of two
modules: (1) an LSTM module for preventing over-
fitting and (2) an LSTM module for forecasting.
In addition, Borovkova and Tsiamas [72] created a
collective structure of LSTM networks to forecast the
intra-day directional changes in stock prices. The
suggested framework used a randomised process for
feature selection and employed RMSProp as the
training algorithm.
 Table 2.0: Previous research work with the use of ML
Authors Technique Accuracy Evaluation Output
Metrics type
MLA Software Feature Data
selection reduction
Chen and FWSVM Not Not Not MAPE, Cat
Hao [45] FWSVM- mentioned mentioned mentioned RMSE
FWKNN
SVM KNN
Chong et al. Deep NN Not PCA, Not Not MAE, Con
[43] mentioned autoencoder, mentioned mentioned NMSE, MI,
and and RMSE
restricted
Boltzmann
Machine
Pimprikar et LR, SVM, NB, Not LR = 82% RMSE Not
al. and mentioned SMV = mentioned
[73] LSTM-NN 60%
Kraus and RNN, NB, Python, bag-of- Not Not Accuracy, Cat: Up
Feuerriegel and LSTM Scikit- words mentioned mentioned MSE, AUC and down
[74] Learn RMSE, and
TensorFlow MAE
and Theano
Thanh et al. Multiple Not PCA PCA Not % absolute Con
[42] regression mentioned mentioned error, R2
Adjusted R2
Adebayo et Takagi- R Not Not Not RMSE, Cat: buy,
al. [75] Sugueno- mentioned mentioned mentioned SMAPE , and sell, or
Kang -Fuzzy MSE hold
Rule-Based
System
Maknickiene RNN Not Not Not 73% Accuracy Cat
et al. [76] mentioned mentioned mentioned
Coyne et al. NB, MLP Python and bag-of- Not 65% Not Not
[77] Scikit- words, LR mentioned mentioned mentioned
Learn
Nayak et al. SVM and Not Not Not Not MAPE, Cat
[78] KNN mentioned mentioned mentioned mentioned RMSE,
MSFE
Boachie et LR STATA Not Not Not R2 p value
al. version 11 mentioned mentioned mentioned
[79]
Wanjawa ANN C# Not Not Not RMSE, and Con
[80] mentioned mentioned mentioned MAPE
Sun et al. Sparse matrix MATLAB Not Not Not Precision Cat
[81] Factorization mentioned mentioned mentioned Recall
Yoosin et al. Natural Not Bag-of- Not Not F1-score, Cat: buy,
[82] Language mentioned words mentioned mentioned Recall, and sell, hold
Processing Precision,

Li, et al. [83] SVR Not Not Not Not RMSEs Con
mentioned mentioned mentioned mentioned
Ballings et RF, LR , SVM Not Not Not Not The portion Cat
al. , NN, Kernel mentioned mentioned mentioned mentioned under the
[84] Factory, K- receiver
NN, and Ada- operating
Boost characteristic
curve (AUC)
Bhagwant et ANN C# Not Not Not MSE Con
al. [85] mentioned mentioned mentioned
Ming et al. Autoregressive Not Bag-of-word Not Not Not Not
[86] (AR) mentioned mentioned mentioned mentioned mentioned
Models
Xi et al. [87] RBF NN MATLAB Not Not Not RMSE Con
mentioned mentioned mentioned
Yetis et al. ANN MATLAB Not Not Not MSE Not
[88] mentioned mentioned mentioned mentioned
 III. RESEARCH GAPS AND ISSUES table 3.0. Typically, a few key measures are often
Despite the availability of a wide variety of approaches used.
to SMP, there are still several hurdles that must be Mean Squared Error (MSE): MSE provides an
overcome before accurate stock market forecasting understanding of the model's overall accuracy by measuring
may be achieved. In this discussion, we will examine the average squared difference between projected values and
the various SMP systems and discuss the areas where actual values. It gives greater weights to bigger mistakes;
more study is needed. however, all errors are measured.
SMP presents the NN with the following difficulties: Mean Absolute Error (MAE): MAE is a simpler metric
However, the neural models cannot tolerate high for prediction errors than MSE as it computes the average
computational cost because of the huge neurons absolute differences between actual and anticipated values.
included in the hidden layer and proper weight Root Mean Squared Error (RMSE): RMSE, which is
adaptation [89], therefore the developed ANN was not measured in the same units as the predicted variable, offers
certified as a successful scheme for forecasting the a statistic that is easily understandable. It exhibits sensitivity
stock market. The prediction performance was to outliers due to its tendency to penalise larger errors more
negatively impacted by the fact that NN, constructed severely than smaller ones.
in [90], took longer to complete both testing and R-squared (R²) or Coefficient of Determination: R² is a
training. Overfitting, being stuck in local minima, and metric that quantifies the degree to which the model
the black box approach are all pitfalls that may be accurately represents the data. It indicates the proportion of
overcome with the help of NN. Due to the impact of the variability in the dependent variable (such as stock
misclassification of the similar patterns and the fact prices) that can be accounted for by the independent
that the network parameters used were not adjusted, variables (such as the predictors in the model).
the results generated by the NN based SMP system Accuracy, Precision, and Recall: The accuracy of models
developed in Pang et al., [91] findings were of poor in making predictions about categories is evaluated using
accuracy. The problem with the study into the CNN- these metrics developed for that purpose. Precision
based stock prediction technique is that the created measures the proportion of accurate positive predictions to
CNN using the DL framework was not well suited for real positives and recall measures the proportion of right
the very broad applications. CNN's recognition predicted positives to all actual positives, whilst accuracy
accuracy was lower than that of other state-of-the-art evaluates the total correctness of forecasts.
prediction systems for stock prediction [92]. The
developed decision support system did not apply the
practical information and skills gained, for creating a
suitable stock expert system, in stock investing [93].
ANN suffers from a lack of explanation for how the
solution is formed, necessitating a lengthy training
procedure for building an ideal model [94]. The
formulation of splitting techniques is quite intricate,
which leads to a complex computing process, and the
fuzzy time series model needed more time and the
accuracy calculated by the fuzzy time series model for
the prediction is continually impacted due to length
[95]. For the purpose of forecasting the direction of
stock markets, SVM implemented a feature selection
strategy. The system's accuracy suffered greatly since
the feature selection approach could not identify the
necessary minimum number of optimum features.
Depending on the feature's correlation value, a SVM
or NN based prediction system is developed [96].
IV. EVALUATION METRICS AND PERFORMANCE ASSESSMENT
Assessing the predictive efficacy of ML techniques
has been an essential component of all the papers in
our review collection. These measurements serve as
the basis for evaluating and comparing the
performance of each algorithm. The measurements
were categorised into four groups: "Accuracy-based",
"Error-based", "Return-driven", and "Statistic tests".
A. Metrics for Evaluating Prediction Models
For assessing the effectiveness and accuracy of
algorithms used for predicting stock market trends, it
is essential to establish assessment criteria as listed in
 Table 3.0: Formualas of Evaluation Metrics
Metric Formula
MSE

MAE

RMSE

R-squared (R2)
1-

Accuracy

Precision

Recall

B. Backtesting and Cross-validation Techniques

methodology they used, the datasets they utilized, their
Backtesting: According to Beaudan [97], a trading or performance measures, and the implementation tools
investing strategy may be "backtested" by simulating its they relied on. ANN, fuzzy-based approaches (K-
prior performance using historical data. It is useful for means), and so on are all used for SMP. For effective
testing trading rules or prediction models in a simulated future scope, the research gaps and concerns with
market before releasing them to the public. stock market forecasting are further explored. The
Cross-validation Techniques: The purpose of cross- generally utilized strategy for getting good SMP is
validation is to evaluate the model's performance across ANN and the fuzzy-based technique. The whole stock
various subsets of the dataset by repeatedly dividing the market may be successfully monitored and controlled
dataset into training and testing subsets [98]. Some popular using these methods. Because of the influence of
approaches that test a model on different data partitions to external variables such as government policy choices,
determine the model's generality and robustness consist of market mood, and so on, most current strategies for
k-fold and leave-one-out cross-validation. predicting stock prices cannot be discovered by
The quantification of accuracy, robustness, and analyzing past stock data. Decisions cannot be made
universality of stock market forecasting models may without information from a variety of sources, and
be achieved by using these evaluation methods and data pre-processing is a challenging aspect of data
methodologies. To ensure the reliability and mining. These are substantial constraints that will have
effectiveness of models prior to their use in real-world to be overcome in the future by modifying established
financial decision-making, it is essential to employ methods for predicting the stock market. We plan to
appropriate metrics and validation techniques. work on perfecting a system of stock market
This table presents the formulas for various evaluation forecasting in the near future.
metrics used in assessing prediction models in a more
structured and readable form.

ACKNOWLEDGMENT

V. CONCLUSION REFERENCES
This study provides a comprehensive overview of the [1] A. Karthick and S. Gopalsamy, “Artificial
many prediction and clustering approaches used to Intelligence: Trends and Challenges,” in 2022
accurately anticipate the stock market. Using 50 Seventh International Conference on Parallel,
studies, this study aims to categorize current methods Distributed and Grid Computing (PDGC), 2022, pp.
according to when they were published, the 7–12, [Online]. Available:
https://doi.org/10.1109/PDGC56933.2022.10053238
 . prediction.,” in 2021 14th International Congress on
[2] T. Tiwari, T. Tiwari, and S. Tiwari, “How Artificial Image and Signal Processing, BioMedical
Intelligence, Machine Learning and Deep Learning Engineering and Informatics (CISP-BMEI), 2021,
are Radically Different?.,” vol. 8, pp. 1–9, 2018, pp. 1–6.
[Online]. Available: [16] Z. Prohaska, I. Uroda, M. Spec, and S. Suljic, “SP
https://doi.org/10.23956/IJARCSSE.V8I2.569. — A computer program for fundamental analysis of
[3] A. Saini and A. Sharma, “Predicting the stocks.,” 2011 Proc. 34th Int. Conv. MIPRO, pp.
Unpredictable: An Application of Machine Learning 1051-1055., 2011.
Algorithms in Indian Stock Market.,” Ann. Data [17] M. Arunagiri, “Fundamental Analysis of Selected
Sci., vol. 9, pp. 791–799, 2019, [Online]. Available: Education Technology Stocks from BSE.,” Int. J.
https://doi.org/10.1007/s40745-019-00230-7. Adv. Res. Sci. Commun. Technol., 2022, [Online].
[4] R. Deshpande, D. Lunkad, M. Kunjir, and S. Ingle, Available: https://doi.org/10.48175/ijarsct-5699.
“Stock Price Prediction and Analysis using Machine [18] V. Pathade, “Equity research: Fundamental analysis
Learning Techniques.,” Int. J. Res. Appl. Sci. Eng. for long term investment.,” Int. J. Appl. Res., vol. 3,
Technol., 2022, [Online]. Available: pp. 678–682, 2017.
https://doi.org/10.22214/ijraset.2022.42747. [19] A. Spritzer and C. Freitas, “A visual tool to support
[5] H. Al-Sahaf et al., “A survey on evolutionary technical analysis of stock market data.,” pp. 512–
machine learning.,” J. R. Soc. New Zeal., vol. 49, 515, 2006, [Online]. Available:
pp. 205–228, 2019, [Online]. Available: https://doi.org/10.1145/1133265.1133372.
https://doi.org/10.1080/03036758.2019.1609052. [20] J. George, A. Nair, and S. Yathish, “Analysis of
[6] M. Sipper, R. Olson, and J. Moore, “Evolutionary Market Behavior Using Popular Digital Design
computation: the next major transition of artificial Technical Indicators and Neural Network.,” Expert
intelligence?,” BioData Min., p. 10, 2017, [Online]. Clouds Appl., 2021, [Online]. Available:
Available: https://doi.org/10.1186/s13040-017- https://doi.org/10.1007/978-981-16-2126-0_37.
0147-3. [21] C. Lohrmann, P. Luukka, M. Jablonska-Sabuka, and
[7] I. Nti, A. Adekoya, and B. Weyori, “A systematic T. Kauranne, “A combination of fuzzy similarity
review of fundamental and technical analysis of measures and fuzzy entropy measures for supervised
stock market predictions.,” Artif. Intell. Rev., vol. feature selection.,” Expert Syst. Appl., vol. 110, pp.
53, pp. 3007–3057, 2019, [Online]. Available: 216–236, 2018.
https://doi.org/10.1007/s10462-019-09754-z. [22] R. Rosillo, J. Giner, and D. L. De Fuente, “The
[8] W. Fraser, “Modern Theories of Investment effectiveness of the combined use of VIX and
Pricing.,” pp. 78–87, 1993, [Online]. Available: support vector machines on the prediction of sandp
https://doi.org/10.1007/978-1-349-13311-6_8. 500.,” Neural Comput. Appl., vol. 25, no. 2, pp.
[9] J. Vesela, “Some Less Known Charting Methods of 321–332, 2014.
Technical Analysis and Possibilities Its Using for [23] E. Hoseinzade and S. Haratizadeh, “CNNPred:
Identification Trend Changes.,” pp. 32–40, 2007, CNN-based stock market prediction using a diverse
[Online]. Available: set of variables.,” Expert Syst. Appl., vol. 129, pp.
https://doi.org/10.18267/J.CFUC.231. 273–285, 2019, [Online]. Available:
[10] L. Yuhan, “Stock Price Prediction based on http://dx.doi.org/10.1016/j.eswa.2019.03.029.
Multiple Regression Models.,” Highlights Sci. Eng. [24] L. S. Malagrino, N. T. Roman, and A. M. Monteiro,
Technol., 2023, doi: 10.54097/hset.v39i.6622. “Forecasting stock market index daily direction: A
[11] Z. Cheng, H. Ping, L. Xiaohui, L. Xuefeng, and Z. Bayesian network approach.,” Expert Syst. Appl.,
Haiming, “Stock Market Analysis Using Time vol. 105, pp. 11–22, 2018, [Online]. Available:
Series Relational Models for Stock Price http://dx.doi.org/10.1016/j.eswa.2018.03.039.
Prediction.,” Mathematics, 2023, doi: [25] S. H. Na and S. Y. Sohn, “Forecasting changes in
10.3390/math11051130. Korea composite stock price index (KOSPI) using
[12] X. Qijun, “Stock Price Forecasting Based on association rules.,” Expert Syst. Appl., vol. 38, no. 7,
ARIMA Model an Example of Cheung Kong pp. 9046–9049., 2011, [Online]. Available:
Hutchison Industrial Co.,” Highlights Business, http://dx.doi.org/10.1016/j.eswa.2011.01.025.
Econ. Manag., 2023, doi: [26] W. Huang, Y. Nakamori, and S. Y. Wang,
10.54097/hbem.v10i.8134. “Forecasting stock market movement direction with
[13] Y. Yue and K. Woohyoung, “Stock Market support vector machine.,” Comput. Oper. Res., vol.
Prediction Based on LSTM Neural Networks.,” 32, pp. 2513–2522, 2005.
Korea Int. Trade Res. Inst., 2023, doi: [27] C. F. Tsai and Y. C. Hsiao, “Combining multiple
10.16980/jitc.19.2.202304.391. feature selection methods for stock prediction:
[14] L. Ruotong, M. Moying, and T. Nan, “Stock Price Union, intersection, and multi-intersection
Prediction Based on Decision Trees, CNN and approaches.,” Decis. Support Syst., vol. 50, no. 1,
LSTM.,” 2023, doi: 10.4108/eai.18-11- pp. 258–269, 2010, [Online]. Available:
2022.2327160. http://dx.doi.org/10.1016/j.dss.2010.08.028.
[15] Y. Lin, Q. Zhong, Q. Huang, M. Li, and F. Ma, “A [28] F. Luo, J. Wu, and K. Yan, “A novel nonlinear
new convolutional neural network and long short combination model based on support vector
term memory combined model for stock index machine for stock market prediction.,” Jinan C
 World Congr. Intell. Control Autom. IEEE, stock market.,” Indian J Sci Technol, vol. 11, no. 2,
Piscataw., pp. 5048–5053, 2010. pp. 1–13., 2018.
[29] S. Agrawal, M. Jindal, and G. Pillai, “Momentum [43] E. Chong, C. Han, and F. Park, “Deep learning
analysis based stock market prediction using networks for stock market analysis and prediction:
adaptive neuro-fuzzy inference system (ANFIS).,” methodology, data representations, and case
2010. studies.,” Expert Syst Appl, vol. 83, pp. 187–205.,
[30] M. Boyacioglu and D. Avci, “Adaptive network- 2017.
based fuzzy inference system (ANFIS) for the [44] Zhou X et al, “Stock market prediction on high
prediction of stock market return: the case of the frequency data using generative adversarial nets.,”
Istanbul stock exchange.,” Expert Syst Appl, vol. 37, Math Probl Eng, vol. 2018, pp. 1–12, 2018.
no. 12, pp. 7908–7912., 2010. [45] Y. Chen and Y. Hao, “A feature weighted support
[31] A. Esfahanipour and W. Aghamiri, “Adapted neuro- vector machine and K-nearest neighbor algorithm
fuzzy inference system on indirect approach TSK for stock market indices prediction.,” Expert Syst
fuzzy rule base for stock market analysis,” Expert Appl, vol. 80, pp. 340–355, 2017.
Syst. Appl., vol. 37, no. 7, pp. 4742–4748, 2010. [46] Y. F. Wang, “Predicting stock price using fuzzy
[32] Y. Kara, B. M. Acar, and Ö. Baykan, “Predicting grey prediction system.,” Expert Syst. Appl., vol. 22,
direction of stock price index movement using no. 1, pp. 33–38, 2002, [Online]. Available:
artificial neural networks and support vector http://dx.doi.org/10.1016/S0957-
machines: the sample of the Istanbul stock 4174(01)%0A00047-1.
exchange.,” Expert Syst Appl, vol. 38, no. 5, pp. [47] R. K. Lai, C. Y. Fan, W. H. Huang, and P. C.
5311–5319., 2011. Chang, “Evolving and clustering fuzzy decision tree
[33] G. Atsalakis, E. Dimitrakakis, and C. Zopounidis, for financial time series data forecasting.,” Expert
“Elliott wave theory and neuro-fuzzy systems, in Syst. Appl., vol. 36, no. 2, pp. 3761–3773, 2009,
stock market prediction: the WASP system.,” Expert [Online]. Available:
Syst Appl, vol. 38, no. 8, pp. 9196–9206, 2011. http://dx.doi.org/10.1016/j.eswa.2008.02.%0A025.
[34] S. Shen, H. Jiang, and T. Zhang, “Stock market [48] T. Anbalagan and S. U. Maheswari, “Classification
forecasting using machine learning algorithms.,” and prediction of stock market index based on fuzzy
Dep. Electr. Eng. Stanford Univ. Stanford, pp. 1–5, metagraph.,” Procedia Comput. Sci., vol. 47, no. C,
2012. pp. 214–221, 2015.
[35] L.-Y. Wei, T.-L. Chen, and T.-H. Ho, “A hybrid [49] R. Efendi, N. Arbaiy, and M. M. Deris, “A new
model based on adaptive-network-based fuzzy procedure in stock market forecasting based on
inference system to forecast Taiwan stock market.,” fuzzy random auto-regression time series model.,”
Expert Syst Appl, vol. 38, no. 11, pp. 13625–13631, Inf. Sci. (Ny)., vol. 441, pp. 113–132, 2018.
2011. [50] R. Arévalo, J. García, F. Guijarro, and A. Peris, “A
[36] J. Anthony, L. Maurice, and S. . Eshwar, “Predictive dynamic trading rule based on filtered flag pattern
ability of the interest rate spread using neural recognition for stock market price forecasting,”
networks.,” Procedia Comput Sci, vol. 6, pp. 207– Expert Syst. Appl., vol. 81, pp. 177–192, 2017.
212, 2011. [51] A. A. Ariyo, A. O. Adewumi, and Ayo, “Stock price
[37] S. Olaniyi, K. Adewole, and R. Jimoh, “Stock trend prediction using the ARIMA model,” Procedings
prediction using regression analysis—a data mining Comput. Model. Simul., pp. 106–112., 2014.
approach.,” ARPN J Syst Softw, vol. 1, no. 4, pp. [52] S. R. Nanda, B. Mahanty, and M. K. Tiwari,
154–157, 2011. “Clustering Indian stock market data for portfolio
[38] E. Guresen, G. Kayakutlu, and T. Daim, “Using management,” Expert Syst. Appl., vol. 37, no. 12,
artificial neural network models in stock market pp. 8793–8798, 2010.
index prediction.,” Expert Syst Appl, vol. 38, no. 8, [53] K. J. Kim, “Financial time series forecasting using
pp. 10389–10397., 2011. support vector machines.,” Neurocomputing, vol.
[39] D. P. Shom and S. Padhy, “Support vector machines 55, pp. 307–319, 2003.
for prediction of futures prices in Indian stock [54] C. Y. Yeh, C. W. Huang, and S. J. Lee, “A multiple-
market.,” Int J Comput Appl, vol. 41, no. 3, pp. 22– kernel support vector regression approach for stock
26, 2012. market price forecasting.,” Expert Syst. Appl., vol.
[40] D. Enke, M. Grauer, and N. Mehdiyev, “Stock 38, no. 3, pp. 2177–2186, 2011, [Online]. Available:
market prediction with multiple regression, fuzzy http://dx.doi.org/10.1016/j.eswa.2010.08.004.
type-2 clustering and neural networks,” Procedia [55] S. P. Das and S. Padhy, “A novel hybrid model
Comput. Sci., vol. 6, pp. 201–206, 2011. using teaching–learningbased optimization and a
[41] K. Sureshkumar and N. Elango, “An efficient support vector machine for commodity futures index
approach to forecast Indian stock market price and forecasting.,” Int. J. Mach. Learn. Cybern., vol. 9,
their performance analysis.,” Int J Comput Appl, no. 1, pp. 97–111, 2018, [Online]. Available:
vol. 34, no. 5, pp. 44–49, 2011. http://dx.doi.org/10.1007/s13042-015-0359-0.
[42] D. V. Thanh, H. N. Minh, and D. Hieu, “Building [56] J. Chai, J. Du, K. K. Lai, and Y. P. Lee, “A hybrid
unconditional forecast model of stock market least square support vector machine model with
indexes using combined leading indicators and parameters optimization for stock forecasting.,”
principal components: application to Vietnamese Math. Probl. Eng., p. 2015, 2015.
[57] P. F. Pai and C. S. Lin, “ARIMA,artificial neural 669, 2018.
networks,stock prices,support vector machines,time [70] H. Chung and K. S. Shin, “Genetic algorithm-
series forecasting.,” Omega, vol. 33, pp. 497–505, optimized long short-term memory network for
2005. stock market prediction.,” Sustainability, vol. 10, p.
[58] C. J. Lu, “Hybridizing nonlinear independent 10, 2018.
component analysis and support vector regression [71] Y. Baek and H. Y. Kim, “ModAugNet: A New
with particle swarm optimization for stock index forecasting framework for stock market index value
forecasting.,” Neural Comput. Appl., vol. 23, no. 7– with an overfitting prevention LSTM module and a
8, pp. 2417–2427, 2013, [Online]. Available: prediction LSTM module.,” Expert Syst. Appl., vol.
http://dx.doi.org/10.1007/%0As00521-012-1198-5. 113, pp. 457–480, 2018, [Online]. Available:
[59] Z. Rustam and P. Kintandani, “Application of http://dx.doi.org/%0A10.1016/j.eswa.2018.07.019.
support vector regression in Indonesian stock price [72] S. Borovkova and I. Tsiamas, “An ensemble of
prediction with feature selection using particle LSTM neural networks for high-frequency stock
swarm optimisation.,” Model. Simul. Eng., 2019, market classification.,” J. Forecast., vol. 38, no. 6,
[Online]. Available: pp. 600–619, 2019, [Online]. Available:
http://dx.doi.org/10.1155/2019/8962717. http://dx.doi.org/10.1002/for.2585.
[60] M. M. Gowthul Alam and S. Baulkani, “Local and [73] R. Pimprikar, S. Ramachadran, and K.
global characteristics-based kernel hybridization to Senthilkumar, “Use of machine learning algorithms
increase optimal support vector machine and Twitter sentiment analysis for stock market
performance for stock market prediction.,” Knowl. prediction.,” Int J Pure Appl Math, vol. 115, no. 6,
Inf. Syst., vol. 60, no. 2, pp. 971–1000, 2019, pp. 521–526, 2017.
[Online]. Available: [74] M. Kraus and S. Feuerriegel, “Decision support
http://dx.doi.org/10.1007/s10115-018-1263-1. from financial disclosures with deep neural
[61] L. Yu, H. Chen, S. Wang, and K. K. Lai, “Evolving networks and transfer learning.,” Decis Support
least squares support vector machines for stock Syst, vol. 104, pp. 38–48., 2017.
market trend mining.,” IEEE Trans. Evol. Comput., [75] A. Adebayo, A. Adekoya, and T. Rahman,
vol. 13, no. 1, pp. 87–102, 2009. “Predicting stock trends using Tsk-fuzzy rule based
[62] L. J. Kao, C. C. Chiu, C. J. Lu, and C. H. Chang, “A system.,” JENRM, vol. 4, no. 7, pp. 48–55, 2017.
hybrid approach by integrating wavelet-based [76] N. Maknickiene, I. Lapinskaite, and A. Maknickas,
feature extraction with MARS and SVR for stock “Application of ensemble of recurrent neural
index forecasting.,” Decis. Support Syst., vol. 54, networks for forecasting of stock market
no. 3, pp. 1228–1244, 2013. sentiments.,” Equilib Q J Econ Econ Policy, vol. 13,
[63] J. L. Ticknor, “A Bayesian regularized artificial no. 1, pp. 7–27, 2018.
neural network for stock market forecasting,” [77] S. Coyne, P. Madiraju, and J. Coelho, “Forecasting
Expert Syst. Appl., vol. 40, no. 14, pp. 5501–5506, stock prices using social media analysis.,” in IEEE
2013. 15th international conference on big data
[64] A. M. More, P. U. Rathod, R. H. Patil, D. R. Sarode, intelligence and computing and cyber science and
and B. E. Student, “Stock market prediction system technology congress., 2017, pp. 1031–1038.
using hadoop, Internat.,” J. Engrg. Sci., p. 16138, [78] R. Nayak, D. Mishra, and A. Rath, “A Naïve SVM-
2018. KNN based stock market trend reversal analysis for
[65] M. R. Vargas, B. S. de Lima, and A. G. Evsukoff, Indian benchmark indices.,” Appl Soft Comput J,
“Deep learning for stock market prediction from vol. 35, pp. 670–680, 2015.
financial news articles,” in Proceedings of IEEE [79] Boachie MK et al, “Interest rate, liquidity and stock
International Conference on Computational market performance in Ghana.,” Int J Acc. Econ
Intelligence and Virtual Environments for Stud, vol. 4, no. 1, p. 46, 2016.
Measurement Systems and Applications, 2017, pp. [80] B. Wanjawa, “Predicting future Shanghai stock
60–65. market price using ANN in the period 21 Sept 2016
[66] X. Zhou, Z. Pan, G. Hu, S. Tang, and C. Zhao, to 11 Oct 2016,” 2016.
“Stock market prediction on high-frequency data [81] A. Sun, M. Lachanski, and F. Fabozzi, “Trade the
using generative adversarial nets,” Math. Probl. tweet: social media text mining and sparse matrix
Eng., 2018. factorization for stock market prediction.,” Int Rev
[67] J. Cao and J. Wang, “Stock price forecasting model Financ Anal, vol. 48, pp. 272–281, 2016.
based on modified convolution neural network and [82] K. Yoosin, R. Seung, and I. Ghani, “Text opinion
financial time series analysis.,” Int. J. Commun. mining to analyze news for stock market
Syst., vol. 32, pp. 1–13, 2019. prediction.,” Int J Adv Soft Comput Appl, vol. 6, no.
[68] M. Wen, P. Li, L. Zhang, and Y. Chen, “Stock 44, pp. 1–13, 2014.
market trend prediction using high-order [83] Q. Li, T. Wang, and Q. et al Gong, “Media-aware
information of time series.,” IEEE Access, vol. 7, quantitative trading based on public Web
pp. 28299–28308., 2019. information.,” Decis Support Syst, vol. 61, no. 1, pp.
[69] T. Fischer and C. Krauss, “Deep learning with long 93–105, 2014.
short-term memory networks for financial market [84] Ballings M et al, “Evaluating multiple classifiers for
predictions.,” Eur. J. Oper. Res., vol. 270, pp. 654– stock price direction prediction.,” Expert Syst Appl,
 vol. 42, no. 20, pp. 7046–7056., 2015. CCF International Conference on Natural
[85] Bhagwant C et al, “Stock market prediction using Language Processing and Chinese Computing,
artificial neural networks.,” Int J Comput Sci Inf 2018, pp. 166–177.
Technol, vol. 5, no. 1, pp. 904–907, 2014. [93] Q. Wen, Z. Yang, Y. Song, and P. Jia, “Automatic
[86] Ming F et al, “Stock market prediction from WSJ: stock decision support system based on box theory
text mining via sparse matrix factorization.,” EEE and SVM algorithm,” Expert Syst. Appl., vol. 37,
Int. Conf. data mining, ICDM, pp. 430–439., 2014. no. 2, pp. 1015–1022, 2010.
[87] Xi L et al, “A new constructive neural network [94] A. Oztekin, R. Kizilaslan, S. Freund, and A. Iseri,
method for noise processing and its application on “A data analytic approach to forecasting daily stock
stock market prediction.,” Appl Soft Comput J, vol. returns in an emerging market,” Eur. J. Oper. Res.,
15, pp. 57–66, 2014. vol. 253, no. 3, pp. 697–710., 2016.
[88] Y. Yetis, H. Kaplan, and M. Jamshidi, Stock market [95] M. Y. Chen and B. T. Chen, “A hybrid fuzzy time
prediction using artificial neural network. ISI Press, series model based on granular computing for stock
2014. price forecasting,” Inform. Sci., vol. 294, pp. 227–
[89] A. K. Rout, B. Biswal, and P. K. Dash, “A hybrid 241, 2015.
FLANN and adaptive differential evolution model [96] L. P. Ni, Z. W. Ni, and Y. Z. Gao, “Stock trend
for forecasting of stock market indices,” Int. J. prediction based on fractal feature selection and
Knowl.-Based Intell. Eng. Syst., vol. 18, no. 1, pp. support vector machine,” Expert Syst. Appl., vol. 38,
23–41, 2014. no. 5, pp. 5569–5576., 2011.
[90] S. Chakravarty and P. K. Dash, “A PSO based [97] P. Beaudan, “Telling the Good from the Bad and the
integrated functional link net and interval type-2 Ugly: How to Evaluate Backtested Investment
fuzzy logic system for predicting stock market Strategies.,” Econom. Model. Cap. Mark. - Asset
indices,” Appl. Soft Comput., vol. 12, no. 2, pp. Pricing eJournal., 2013, [Online]. Available:
931–941., 2012. https://doi.org/10.2139/ssrn.2346600.
[91] X. Pang, Y. Zhou, P. Wang, W. Lin, and V. Chang, [98] T. Gunasegaran and Y. Cheah, “Evolutionary cross
“An innovative neural network approach for stock validation.,” in 2017 8th International Conference
market prediction,” J. Supercomput., pp. 1–21., on Information Technology (ICIT), 2017, pp. 89-95.,
2018. [Online]. Available:
[92] B. Xu, D. Zhang, S. Zhang, H. Li, and H. Lin, https://doi.org/10.1109/ICITECH.2017.8079960.
“Stock market trend prediction using recurrent
convolutional neural networks,” in Proceedings of