PREDICTIVE MODELING FOR SUGARCANE PRODUCTION: A COMPREHENSIVE COMPARISON OF ARIMA AND MACHINE LEARNING ALGORITHMS

Accurate prediction of sugarcane yield is essential for trade, economic planning, and sustainable agriculture in India. This study addressed the challenge of forecasting sugarcane yield by evaluating the effectiveness of time series modelling and machine learning algorithms. Leveraging data spanning from 2001 to 2020, the research focuses on predicting the sugarcane yield for the subsequent years. The problem statement revolves around the need for precise yield predictions to inform decision-making in the agricultural sector. Methods employed included the utilization of Autoregressive Integrated Moving Average (ARIMA) for time series analysis and machine learning algorithms such as Random Forest (RF), Support Vector Machine (SVM), and Gradient Boosting Machine (GBM). The analysis encompassed sugarcane yield data spanning multiple years, with predictions extending for a specified duration. Through analysis of temporal patterns and dependencies within the sugarcane yield time series data using Autocorrelation Function (ACF) and Partial Autocorrelation Function (PACF), the study optimized the predictive models. Results indicated that ARIMA outperformed machine learning algorithms, exhibiting superior performance with a root mean square error of 36700.68 and a minimum AIC value of 456.7. The study emphasizes the significance of accurate yield predictions for agricultural planning and decision-making, highlighting the implications for sustainable crop management and the fortification of Indian sugar industry. The study affirms the importance of informed decisions facilitated by accurate yield predictions in resilient agricultural sector. Overall, this study contributes to the advancement of sugarcane yield prediction, offers practical insights for stakeholders and policymakers in India's agricultural landscape.