Evaluating Statistical and Machine Learning Techniques for Sugarcane Yield Forecasting in the Tarai Region of North India

Sugarcane holds paramount importance as a key raw material in India’s second-largest agro-based industry, particularly in the state of Uttar Pradesh, which boasts the largest sugarcane cultivation area. Amidst unprecedented climate changes, accurate crop yield forecasting is imperative for informed decision-making in policy formulation, marketing, and pricing. Looking at the facts, eleven major sugarcane producing districts of Uttar Pradesh located in tarai belt viz. Bahraich, Bareilly, Basti, Bijnor, Deoria, Gonda, Gorakhpur, Lakhimpur, Pilibhit, Rampur and Saharanpur were selected for sugarcane yield forecasting. Weather and yield data encompassing the sugarcane yield spanning 22 years, from 1998 to 99 to 2019–20, were gathered for all mentioned districts. Out of the total datasets 75 % were utilized for model training, while the remaining 25 % were reserved for model testing. This study adopts diverse methods, including multiple linear regression (MLR), three penalized regression techniques viz. Least absolute shrinkage selection operator (LASSO), Ridge regression, Elastic Net (ELNET) and four advanced machine learning approaches. Extreme gradient boosting (XGB), Random Forest (RF), Support vector regression (SVR) and Artificial neural network (ANN). Contrary to expectations, the findings indicate that penalized regression models do not significantly enhance sugarcane yield prediction compared to the traditional regression approach. Machine learning methods emerge as more promising alternatives, with Artificial Neural Network (ANN) demonstrating superior performance. Ridge regression and MLR exhibit poor performance in sugarcane yield prediction compared to other models in the study regions. The study suggests that ANN can be a reliable method for accurate sugarcane yield forecasting across all specified districts.