Predicting Agriculture Yields Based on Machine

Learning Using Regression and Deep Learning

Doddikindi Navya*, Logaom Ajay**, Nagulapally Nithish Reddy***, Dr. A.Ramesh Babu

* Information Technology
** J.B.Institute of Engineering and Technology

Abstract- Agriculture plays a crucial role in India's enhance productivity and mitigate risks associated with
economy, serving as the backbone of the nation's climate change and resource constraints. Future
livelihood and food security. However, rapid population advancements may include incorporating real-time data,
growth has significantly increased the demand for food, satellite imagery, and IoT-based sensors to refine
creating pressure on agricultural production. To meet these predictive accuracy further. The findings of this research
rising demands, farmers must enhance crop yields without contribute to the ongoing efforts to revolutionize
expanding cultivable land. Technology-driven solutions, agriculture through data-driven methodologies, ensuring a
particularly machine learning and deep learning, offer more sustainable and efficient food production system. By
promising approaches to addressing this challenge by leveraging these innovative technologies, India can take
optimizing agricultural output through accurate crop yield significant strides toward achieving long-term agricultural
predictions. resilience and food security.
Crop yield prediction serves as a valuable decision-support
tool, leveraging advanced computational techniques to
analyze factors such as rainfall, meteorological conditions, Index Terms- Agriculture,Crop Yield Prediction, Food
soil quality, cultivated area, production trends, and yield Security, Machine Learning, Deep Learning, Random
history. By utilizing machine learning and deep learning Forest, Convolutional Neural Network (CNN),Long Short-
models, farmers and policymakers can make informed Term Memory (LSTM), Decision Tree, XGBoost
decisions regarding crop selection, resource allocation, and Regression, Meteorological Conditions,
farming practices, thereby improving agricultural Rainfall,Soil,Mean Absolute Error (MAE), Root Mean
sustainability and reducing yield losses due to Square Error (RMSE), Mean Squared Error (MSE),
environmental uncertainties. Standard Deviation, Loss Function, IoT-based Sensors,
This study aims to develop an effective crop yield Satellite Imagery, Predictive Analytics, Data-driven
prediction model using machine learning algorithms such Agriculture
as Decision Tree, Random Forest, and XGBoost I. INTRODUCTION
regression, alongside deep learning approaches, including Agriculture plays a fundamental role in human civilization,
Convolutional Neural Networks (CNN) and Long Short- not only as a primary source of food but also as a key
Term Memory (LSTM) networks. These models are contributor to employment and economic stability. While
evaluated based on key performance metrics such as humans have been consuming grains and plants for over
accuracy, Root Mean Square Error (RMSE), Mean 100,000 years, systematic crop cultivation and land
Absolute Error (MAE), Mean Squared Error (MSE), management emerged approximately 11,000 years ago
standard deviation, and loss functions. during the Neolithic era, commonly known as the New
Comparative analysis reveals that the Random Forest Stone Age. In India, agriculture remains a crucial economic
algorithm outperforms other machine learning methods, driver, fulfilling the majority of the nation’s food
achieving a maximum accuracy of 98.96%, a Mean requirements and employing a significant portion of the
Absolute Error of 1.97, an RMSE of 2.45, and a standard workforce.
deviation of 1.23. Meanwhile, among deep learning However, India's rapid population growth and evolving
models, the Convolutional Neural Network demonstrates climatic conditions pose substantial challenges to
superior performance with a minimum loss of 0.00060. maintaining stable food supply chain. To address these
These results indicate that both Random Forest and CNN challenges, agritech innovations and data-driven farming
are highly effective in predicting agricultural yield, techniques have been integrated into the agricultural sector.
offering robust insights for farmers and stakeholders. Fluctuating weather patterns, irregular rainfall, and land-
Furthermore, the study underscores the importance of use constraints make it difficult for farmers to adopt
integrating machine learning into agricultural practices to sustainable and resilient agricultural practices. With the
advent of precision agriculture and smart farming, improving forecasting accuracy. Classification models,
technology-driven solutions can help maximize yield with image recognition, and computer vision technologies are
minimal resource input. also emerging as powerful tools in crop monitoring,
As agriculture utilizes approximately 70% of the world's disease detection, and fruit classification within modern
freshwater resources, ensuring optimal productivity is vital. agriculture.
In 2018, over 50% of India’s workforce was engaged in This study aims to develop a robust yield prediction
agriculture, contributing 17%-18% of the national GDP. framework using historical agricultural datasets to provide
Despite its economic importance, traditional farming early forecasts that can help farmers make informed
methods often lack the efficiency needed to meet decisions, prevent financial losses, and ultimately enhance
increasing food demands. The Indian agricultural industry national food security. Predicting agricultural yield is
is projected to grow to USD 24 billion by 2025, with the complex, as multiple variables—including rainfall, wind
country ranking 6th in the global food and grocery market. speed, soil health, climate, humidity, and temperature—
As per preliminary estimates for the 2022-2023 fiscal year, impact production. No single dataset comprehensively
India's total food grain production is expected to reach captures all influencing factors, requiring data collection
149.92 million tons, driven primarily by Kharif crops. from multiple sources.Although numerous studies have
explored crop yield estimation, achieving superior
Higher urban and rural income levels have also contributed predictive performance remains a research priority. The
to rising demand for agricultural produce. To bridge the integration of machine learning and deep learning models
supply-demand gap, agricultural digitalization is gaining presents a promising avenue for enhancing forecasting
momentum. Emerging technologies such as Geographic accuracy and minimizing yield losses.
Information Systems (GIS), Artificial Intelligence (AI),
Blockchain, Remote Sensing, and Drones are reshaping the
agricultural landscape. Figure 1 illustrates seasonal crop II. RESEARCH AND IDEA
production trends, highlighting that Kharif season
experiences the highest yield, while winter season reports This research focuses on utilizing Machine Learning (ML)
the lowest. and Deep Learning (DL) techniques to predict agricultural
In recent years, ML-driven models such as Decision Trees, yield by analyzing various factors such as rainfall, crop
Artificial Neural Networks (ANNs), Support Vector type, meteorological conditions, land area, and production
Machines (SVMs), and Deep Learning frameworks have trends. With India's rapidly growing population and
been extensively applied to predict agricultural output. increasing climate variability, accurately estimating crop
Figure 2 presents an analysis of India's crop production yields is essential for ensuring food security and economic
trends from 1997 to 2020, showcasing that wheat and rice stability. By integrating advanced technologies, farmers
dominate the cultivation landscape, contributing to over can optimize crop selection, resource management, and
73% of the country’s staple grain production. decision-making processes to enhance productivity and
India holds a 40% share in the global rice trade, exporting reduce losses.
Basmati and Non-Basmati rice to over 150 countries. In the
first half of 2022-2023, exports surged by 11% to 2.16 The study employs a combination of ML and DL models,
million tons, reinforcing India's dominance in the global where Random Forest (RF) and Convolutional Neural
agricultural market. Figure 3 highlights India's rice export Networks (CNN) are identified as the best-performing
trends from 2015-2022, demonstrating consistent growth in models for crop yield forecasting. To achieve this,
international trade. historical data on rainfall, temperature, humidity, soil
Throughout history, agricultural practices have evolved conditions, crop yield, and area-wise production is
from rudimentary astronomical observations and religious collected from official sources. The dataset undergoes
rituals to advanced scientific methodologies. The Industrial preprocessing, and key agricultural parameters are selected
Revolution introduced mechanization, mathematical for analysis. Several predictive models, including Decision
modeling, and standardized measurement tools, Tree, Random Forest, and XGBoost regression, are trained
accelerating the transition to data-driven decision-making. and evaluated, along with deep learning approaches such as
By the 20th century, regression analysis became a widely CNN and Long Short-Term Memory (LSTM) networks.
used statistical approach to assess agricultural productivity The study uses k-fold cross-validation to ensure robust
by examining the relationships between weather patterns, model performance.
soil conditions, and historical yield data.
For nearly a century, predictive modeling has played a The findings indicate that Random Forest achieves an
pivotal role in farming, with recent advances in accuracy of 98.96%, with a Mean Absolute Error (MAE)
computational power and data accessibility significantly
of 1.97 and a Root Mean Square Error (RMSE) of 2.45,
making it the most effective ML model. Meanwhile,
among deep learning models, CNN outperforms others
with a minimum loss of 0.00060, demonstrating its
robustness in predicting agricultural yield. The study
emphasizes how integrating ML and DL techniques can
significantly enhance forecasting accuracy, helping farmers
make informed decisions about crop planning and resource
allocation.

Looking ahead, future advancements can further improve

prediction accuracy by incorporating IoT sensors, satellite
imagery, and remote sensing data into the model. These
innovations can support climate-resilient farming by
providing real-time insights, enabling farmers to adapt to
environmental changes and maximize yield with minimal
resource consumption. Additionally, the research
highlights the potential for expanding the model to other
regions, thereby contributing to global food security and
sustainable agricultural practices. By leveraging AI-driven
insights, this study lays the groundwork for precision
agriculture, reducing economic losses and ensuring an
efficient and resilient food production system.
Figure1: A detailed infographic showing agricultural data flow

III. SCOPE OF THE PROJECT

farmers.Enhancing the model with real-time climate and
market price forecasting.
The scope of this project encompasses the development
and application of Machine Learning (ML) and Deep IV. THE PROPOSED SYSTEM
Learning (DL) models to predict agricultural yields with
The proposed system for agricultural yield prediction is an
high accuracy. The project aims to assist farmers,
AI-driven, data-centric framework that integrates Machine
policymakers, and agricultural industries in making data-
Learning (ML), Deep Learning (DL), and Internet of
driven decisions to improve productivity, resource
Things (IoT) technologies to enhance crop yield
management, and food security.Analyzing historical crop
forecasting accuracy. This system aims to empower
yield data with factors like rainfall, temperature, soil
farmers, agricultural analysts, and policymakers with real-
quality, and land area. Implementing ML and DL models
time, predictive insights for optimized decision-making in
such as Random Forest, XGBoost, CNN, and LSTM for
farming and resource management.
yield forecasting. Enhancing accuracy and reliability of
1. Data Acquisition & Preprocessing
yield predictions for various crops. Integrating IoT-based
The system collects multi-source agricultural data from:
sensors, satellite imagery, and remote sensing data for real-
Satellite-based remote sensing for real-time climate
time monitoring.Providing early warnings for low-yield
monitoring.IoT-enabled soil sensors for pH levels,
risks due to environmental changes. Assisting farmers in
moisture content, and nutrient concentration.
optimal resource allocation (fertilizers, pesticides,
Historical meteorological records for rainfall patterns,
irrigation).
temperature fluctuations, and wind speed.
Reducing food wastage and overproduction losses by Geospatial crop mapping to analyze land use patterns and
forecasting demand accurately.Supporting climate-resilient vegetation health.
farming practices through predictive analytics.Improving After collection, the data undergoes standardization,
water and soil management to ensure sustainable feature selection, and noise reduction using advanced data
agricultural practices.Expanding the model to different engineering techniques to ensure accuracy and
geographical regions and crop varieties.Developing a consistency.The core of the proposed system involves
mobile or web-based decision-support system for hybrid AI algorithms that combine:
Ensemble Learning Models (Random Forest, XGBoost) for V. RESULTS
structured data analysis.
Temporal Sequence Models (LSTM, Transformer The dataset features exhibit strong interconnections,
Networks) for time-series forecasting of crop yields. with crop type emerging as a crucial variable
Deep Convolutional Networks (CNNs) for processing influencing production. The authors identified and
satellite imagery and soil texture classification. visualized this relationship, highlighting production
Reinforcement Learning (RL) for adaptive crop
counts for ten selected trends in land allocation across
recommendation systems based on climate variability and
soil fertility.
different crop categories. Notably, wheat occupies the
largest cultivated area, followed by rice.These
visualizations emphasize historical dependencies
between key agricultural parameters such as crop
type, land area, and total yield.

Demonstrates that the Random Forest algorithm

outperforms other machine learning models in terms
of predictive accuracy. Statistical evaluations indicate
that Random Forest delivers India's most precise crop
yield estimations, achieving an accuracy of 98.96%,
with a Mean Absolute Error (MAE) of 1.97, Root
Mean Square Error (RMSE) of 2.45, and a
Figure 2: Flow chart of the approach used. Standard Deviation (SD) of 1.23.In contrast,
alternative models such as Decision Tree and
Decision Support & Visualization The system features an XGBoost exhibit lower predictive accuracy. The
interactive AI dashboard that provides Dynamic yield Decision Tree model records an accuracy of 89.78%
forecasting reports with real-time updates.Heatmaps and with MAE = 4.58, RMSE = 5.86, and SD = 2.75,
trend analysis for crop growth and climate whereas XGBoost achieves 86.46% accuracy, MAE
influence.Predictive alerts for potential risks such as
= 6.31, RMSE = 7.89, and SD = 3.54. Figures 13, 14,
droughts, floods, and pest infestations. Automated crop
selection advice based on regional climate, soil conditions, and 15 provide a comparative evaluation of these
and market demands. models' predictive performances.
Deployment & Scalability
The final system will be deployed as a cloud-based
platform accessible via:Mobile applications for farmers to
receive real-time insights. Web-based dashboards for
agribusinesses to analyze large-scale crop trends.APIs for
government agencies to integrate with agricultural policies
and food security programs.
With scalability in mind, the system will continuously
improve predictions by integrating reinforcement learning,
real-time satellite feeds, and blockchain-based farm data
authentication for trustworthy and tamper-proof
agricultural analytics.
Expected Impact Enhanced productivity by optimizing
farming strategies using AI-driven predictions.Climate-
adaptive agriculture with real-time risk mitigation
strategies. Improved food security through intelligent
supply chain forecasting. Sustainable farming by reducing Figure 3: Model performance of CNN.
water consumption, fertilizer overuse, and crop wastage.
This next-generation agricultural yield prediction system Machine learning methodologies, often considered “black
aims to revolutionize smart farming, ensuring resilience box” approaches due to their lack of interpretability,
and efficiency in global food production. demonstrate varying levels of efficacy in agricultural yield
forecasting. Among them, Random Forest emerges as the
most reliable model, surpassing other regression-based are applied to the dataset taken into consideration. When
techniques. data is analyzed at the country level, Random Forest
The deep learning models CNN (Convolutional Neural VOLUME 11, 2023 P. Sharma et al.: Predicting
Networks) and LSTM (Long Short-Term Memory Agriculture Yields Based on Machine Learning (with
Networks) were also examined for performance. Figure accuracy-98.96%, mean absolute error-1.97, RMSE 2.45
16 outlines the architectural layers utilized in these models. and standard deviation-1.23) and CNN (with minimum
Based on Table 2, the test losses for CNN and LSTM loss-0.00060) perform better according to the current
were recorded as 0.00060 and 0.00063, respectively. prediction. Experimental findings demonstrate that the
Performance evaluations (Figures 17 and 18) indicate that approach has a great potential for precise crop productivity
CNN exhibits superior efficiency over LSTM, attributed prediction and its effectiveness has been validated using
to its lower loss function values. real-time data and interactions with people. More data for
each crop year having more historically precise
information about the climate and environment is needed.
More deep learning models need to be applied to the
dataset to identify the method that performs the best. To
increase the model’s accuracy in crop production
prediction, remote sensing data could be amalgamated with
statistical data of districts. The prediction can be more
accurate using satellite imagery land cover or satellite
image classification

REFERENCES
1. C.H.Vanipriya, Maruyi, S. Malladi, and G. Gupta, ‘‘Artificial
intelligence enabled plant emotion expresser in the development
hydroponics system,’’ Mater. Today, Proc., vol. 45, pp. 5034–5040,
Jan. 2021. [21]
2. A.Tomar,G.Gupta,W.Salehi, C.H. Vanipriya, N.Kumar, and
Figure 2:CNN v/s LSTM. B.Sharma, ‘‘A review on leaf-based plant disease detection systems
using machine learning,’’ in Proc. ICRIC, vol. 1, 2022, pp. 297–303.
[22]
A comparative assessment of CNN and LSTM reveals 3. Govt India. (2023). Profile. Accessed: Jan. 20, 2023. [Online].
that variations in the number of training epochs Available: https://www.india.gov.in/india-glance/profile [23]
4. Govt India. (2023). Data. Accessed: Jan. 20, 2023. [Online].
significantly impact predictive accuracy. The results Available: https://data.gov.in [24]
suggest that CNN is the preferred deep learning model due 5. Govt India. (2023). Crop Production Statistics Information System.
to its lower error rate, offering enhanced precision in crop Accessed: Jan. 20, 2023. [Online]. Available: https://aps.dac.gov.
in/APY/Index.htm [25]
yield forecasting compared to LSTM. 6. D. J. Reddy and M. R. Kumar, ‘‘Crop yield prediction using machine
learning algorithm,’’ in Proc. 5th Int. Conf. Intell. Comput. Control
Syst. (ICICCS), May 2021, pp. 1466–1470. [26]
7. S. Bhansali, P. Shah, J. Shah, P. Vyas, and P. Thakre, ‘‘Healthy
harvest: Cropprediction and diseasedetection system,’’ in Proc.
IEEE7thInt.Conf. Converg. Technol. (I2CT), Apr. 2022, pp. 1–5. [27]
VI. CONCLUSION 8. S. Agarwal and S. Tarar, ‘‘A hybrid approach for crop yield prediction
using machine learning and deep learning algorithms,’’ J. Phys., Conf.
The demand and supply for food have grown more difficult Ser., vol. 1714, no. 1, Jan. 2021, Art. no. 012012.
to manage as the population grows. To assist farmers,
experts have worked hard over the past few years to AUTHORS
anticipate agricultural yield production. In order to forecast First Author – Doddikindi Navya, B.Tech(IT) JBIET and
India’s crop yield, this study uses various machine learning [email protected]
and deep learning approaches. The study underlines the Second Author – Logaom Ajay, B.Tech(IT) JBIET and
advantages of cutting-edge procedures. It is beneficial for [email protected]
small-scale ranchers, as they may use the predictions to Third Author – Nagulapally Nithish Reddy, B.Tech(IT) JBIET
estimate crop and [email protected]
productionforupcomingyearsandplantitappropriately.Five Internal Guide – Dr.A. Ramesh Babu Sir, Asst.Proff &HOD
machine learning and deep learning algorithms, Decision (IT),JBIET, and [email protected]
Tree, Random Forest, XGBoost regression, Convolutional
Neural Network, and Long-Short Term Memory Networks