Application of artificial neural networks for rice and wheat crop yield prediction in Chhattisgarh

Introduction

Climate variability poses significant challenges to agricultural productivity, necessitating accurate and robust crop yield prediction models to support sustainable farming practices [Pooja et al. (2025)] [11] . As climate anomalies continue to increase in frequency and intensity, the need for accurate crop yield predictions becomes imperative for the assurance of food security and the optimization of resource management [Ray et al. (2015)] [12] . Crop yield prediction is one of the most difficult tasks in precision agriculture. The ability to forecast crop yields enables the relevant authorities to make appropriate decisions to ensure food security. In addition to soil, genotype, and management techniques, weather conditions have a significant impact on crop yield [Singh et al. (2014)] [14] . Traditional forecasting methods frequently face significant challenges in accurately representing the intricate relationships between climate variables and agricultural productivity [Chlingaryan et al. (2018)] [6] . The comparative analysis revealed that, among the evaluated models, Multiple Linear Regression, Ridge Regression, and Artificial Neural Network, the ANN consistently outperformed the traditional approaches, achieving the highest predictive accuracy and demonstrating superior capability in capturing complex, non-linear relationships between climatic variables and crop yields [Satpathi et al. (2023)] [13] . However, further research is needed in crop yield prediction. The primary objective of this study is to evaluate the effects on crop yield prediction through the implementation of ANN, MLR, and Ridge Regression. This study analyzes historical climate and yield data to determine which model can provide the most accurate crop yield predictions.

Materials and Methods

Study Area

This study examines crop yield prediction in the Chhattisgarh Central Plains, a major agricultural region of India known for rice and wheat production. Covering 1, 35, 194 sq. km, the area's farming is shaped by diverse soils, fluctuating temperatures, variable rainfall, and irrigation from rivers (Mahanadi, Shivnath, Hasdeo, Arpa) and canals. Geographically located between 17°46'-24°06' N latitude and 80°15'-84°20' E longitude, the region has three main agricultural zones: Raipur, Bilaspur, and Durg, which were later subdivided into smaller districts. For analysis, historical weather and crop yield data were consolidated at the regional level to maintain continuity. These regions were chosen for their agricultural importance, data availability, and climatic variability. The research utilized a dataset collected from various websites between 1997 and 2023. These sources included the Directorate of Economics and Statistics (DES), ICRISAT-District level data, and the meteorological website https://power.larc.nasa.gov.

Methodology

In this study, crop yield prediction was evaluated using machine learning (ML) model as well as traditional models. We collected historical climate data and crop yield records and performed data preprocessing to handle missing values, normalize continuous variables, and identify key climatic indicators affecting crop growth. Three models, ANN, MLR, and Ridge Regression, were implemented using R software. The dataset was divided into training (80%) and testing (20%) subsets to ensure an unbiased evaluation. Model performance was assessed using statistical metrics such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and the coefficient of determination (R²).

Models 2.3.1 Artificial Neural Network

Artificial Neural Networks are computational models inspired by the biological neural networks of the human brain. The human brain consists of billions of neurons working together as a highly complex and efficient network. Similarly, ANNs are made up of interconnected artificial neurons (nodes) that process data collaboratively. These models are particularly useful for solving problems involving pattern recognition, classification, and prediction. The architecture of ANNs allows them to capture complex and nonlinear relationships in data, distinguishing them from traditional statistical methods. An ANN typically consists of three types of layers: an input layer, one or more hidden layers, and an output layer. The input layer receives external data that represents the features or variables of interest. The hidden layers contain neurons that perform computations on the input data and learn underlying patterns. Networks with multiple hidden layers are called "deep" neural networks. Finally, the output layer provides the network's decision or prediction. Each connection between neurons has an associated weight that signifies the strength or influence of the input on the network's output. ANN has its ability to learn from data. The learning process generally involves two phases: forward propagation and backpropagation. In forward propagation, the input data passes through the network, layer by layer, until an output is generated. The predicted output is then compared with the actual result, and the error is calculated. During backpropagation, this error is sent backward through the network, and the connection weights are adjusted to minimize the error. This cycle repeats multiple times, allowing the network to gradually improve its accuracy and make better predictions.

Multiple Linear Regression

Multiple regression analysis involves incorporating several independent variables simultaneously to estimate the impact on a dependent variable (Snedecor and Cochran, 1967). In this study, the multiple linear regression equation was developed with crop yield as the dependent variable and various weather parameters as the independent variables. The generalized representation of the multiple regression model is provided below. Y = β0 + β1X1 + β2X2+ …………. + βnXn+ ϵ In the regression model, Y represents the dependent variable, while β₀ denotes the intercept. The term βᵢ corresponds to the regression coefficient of the ith independent parameter (i = 1, 2, …, n), and Xᵢ represents the ith weather parameter. The model also includes an error term, ϵ, which accounts for unexplained variability in the dependent variable.

Ridge Regression

Ridge Regression, also known as Tikhonov regularization, is a traditional statistical model and a modified form of Multiple Linear Regression. It improves model stability and prediction accuracy by shrinking the coefficients toward zero. This method is especially useful when explanatory variables suffer from multicollinearity, a condition where predictors are highly correlated with each other. In such cases, the coefficients estimated by Ordinary Least Squares become unstable and sensitive to small data changes. Ridge Regression addresses this by adding a penalty term to the regression objective function. Similar to OLS, Ridge Regression minimizes the Residual Sum of Squares (RSS), but it adds a tuning parameter λ (lambda) as a penalty to control coefficient size. The objective function is:

In the model, yᵢ represents the dependent variable or response variable, while xᵢⱼ denotes the j-th predictor variable for the ith observation. β₀ is the intercept term, and βⱼ is the coefficient corresponding to predictor j. The total number of predictors is represented by p, and the number of observations is n. Additionally, λ is the tuning parameter that controls the size of the penalty applied in the model.

Model accuracy evaluation:

The coefficient of determination (R²) employed in this study to evaluate the extent to which the explanatory variables align with crop production outcomes. The precision of the crop production estimations was assessed through the use of RMSE and MSE.

Where, N stands for the number of data, yi for the actual yield value and yˆi for the predicted model value. For optimal model performance, it is crucial that R² is maximized and RMSE, MSE and MAE are minimized.

Results and Discussion

This section presents the results of three models ANN, MLR and Ridge Regression in predicting rice and wheat yield in the central plains of Chhattisgarh using historical weather and crop yield data. The dataset was split into 80% for training and 20% for testing to ensure unbiased evaluation. Key performance metrics such as RMSE, MSE and R² were used to assess model effectiveness. Before implementation, we will load and preprocess the climate and crop yield datasets. Relative humidity has a negative correlation with rice but a positive one with wheat. Lastly, wind speed negatively impacts both crops, more so for rice. These correlations help in understanding how the weather influences crop yields. that is explained by the independent variables (Weather parameters); a higher R² value indicates better explanatory power. MSE is the average of the squared differences between actual and predicted values, whereas RMSE, being the square root of MSE, expresses the error in the original unit, making it more interpretable. In this study, 27 years (1997-2023) of weather and yield data, including parameters such as temperature, rainfall, humidity, and wind speed, were used to predict the yield of rice and wheat. ANN, a machine learning-based non-linear model, was found capable of capturing complex relationships, while MLR and Ridge Regression, as standard statistical approaches, offered the benefits of interpretability, stability, and lower computational complexity. Ridge Regression employs regularization to mitigate multicollinearity, whereas ANN achieved higher prediction accuracy but required longer training time and careful parameter tuning. The entire analysis and modeling were performed using R software, with 80% of the data used for training and 20% for testing. The results are based on the historical data and climatic conditions of the central plains of Chhattisgarh. Ridge Regression showed moderate success, while MLR struggled to reflect the area's diverse agricultural conditions. This makes ANN highly suitable for planning and decisionmaking in the central plains.

Predicting crop yield using the artificial neural network

Conclusion

The study confirms that the ANN offers the most reliable and precise predictions for rice and wheat yields in Chhattisgarh's central plains. Its ability to handle complex non-linear relationships and respond to varying weather conditions allows it to consistently outperform traditional models. MLR and Ridge Regression were useful in identifying general trends but were limited by their linear frameworks, restricting their capability to account for multidimensional climate yield interactions. ANN's adaptability, capacity to manage large datasets, and iterative learning process provide it with a clear edge, making it a powerful instrument for agricultural planning, resource optimization, and climate risk mitigation.

Incorporating ANN into precision agriculture can greatly enhance crop management and help build resilient, sustainable food systems.

References

1. Agrawal R, Singh KN, Lama A, Gurung B, Haque MA, Singh KK, et al. Weather based models for pre-harvest crop yield forecasting. J Indian Soc Agric Stat. 2023;77(1):59-69.
2. Aravind KS, Vashisth A, Krishanan P, Das B. Wheat yield prediction based on weather parameters using multiple linear, neural network and penalised regression models. J Agrometeorol. 2022;24(1):18-25.
3. Bejo SK, Mustaffha S, Ismail WI. Application of artificial neural network in predicting crop yield: A review. J Food Sci Eng. 2014;4:1-9.
4. Bhojani SH, Bhatt N. Wheat crop yield prediction using new activation functions in neural network. Neural Comput Appl. 2020;32(17):13941-13951.
5. Champaneri M, Chachpara D, Chandvidkar C, Rathod M. Crop yield prediction using machine learning. Int J Sci Res. 2020;9(4):645-648.
6. Chlingaryan A, Sukkarieh S, Whelan B. Machine learning approaches for crop yield prediction and nitrogen status estimation in precision agriculture: A review. Comput Electron Agric. 2018;151:61-69.
7. Das B, Nair B, Reddy VK, Venkatesh P. Evaluation of multiple linear, neural network and penalised regression models for prediction of rice yield based on weather parameters for west coast of India. Int J Biometeorol. 2018;62(10):1809-1822.
8. Demirel MH, Baran MF, Gokdogan O. Effect of input usage on wheat yield: An application of artificial neural networks. J Agric Sci Technol. 2024;26(2):122-134.
9. Dhekale BS. Statistical models for prediction of crop yield. J Indian Soc Agric Stat. 2014;68(3):331-338.
10. Kakati N, Deka RL, Das P, Goswami J, Khanikar PG, Saikia H. Forecasting yield of rapeseed and mustard using multiple linear regression and ANN techniques in the Brahmaputra valley of Assam, North East India. Theor Appl Climatol. 2022;150:1201-1215.
11. Pooja, Kumar PS, Ramole S, Narsimhaiah L, Saxena RR. Impact of climate variability on groundnut yield prediction using AI models in Chhattisgarh. Int J Agric Stat Sci. 2025;21(1):237-243.
12. Ray DK, Gerber JS, MacDonald GK, West PC. Climate variation explains a third of global crop yield variability. Nat Commun. 2015;6:5989.
13. Satpathi A, Setiya P, Das B, Nain AS, Jha PK, Singh S, et al. Comparative analysis of statistical and machine learning techniques for rice yield forecasting for Chhattisgarh, India. Sustainability. 2023;15(3):2786.
14. Singh RS, Patel C, Yadav MK, Singh KK. Yield forecasting of rice and wheat crops for eastern Uttar Pradesh. J Agrometeorol. 2014;16:199-202.
15. Verma OP, Pandey D, Singh P. Evaluation of multiple linear, neural network and penalised regression models for prediction of rice yield based on weather variables. Int J Agric Stat Sci. 2016;12(1):123-129.