Machine learning-based forecasting of cassava production for smart agricultural planning in India

Reliable forecasts of agricultural production are essential for improving crop planning, resource allocation, and policy decisions in modern farming systems. Cassava (Manihot esculenta Crantz) is an important tropical root crop contributing to food security and agro-processing industries in India. This study evaluates statistical and machine learning approaches for forecasting cassava cultivated area, production, and productivity using annual data from 2000 to 2024. Three forecasting models were applied: Autoregressive Integrated Moving Average (ARIMA), Random Forest Regression (RFR), and Time Delay Neural Network (TDNN). Stationarity was verified using the Augmented Dickey–Fuller test, and the dataset was divided into training and testing sets. Forecasting accuracy was assessed using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). The TDNN model outperformed ARIMA and RFR, recording RMSE values of 3.65, 377.82, and 2.34; MAE values of 2.98, 341.26, and 2.18; and MAPE values of 3.89%, 5.03%, and 5.59% for cultivated area, production, and productivity, respectively. Forecasts for 2025–2030 indicate stable cultivated area, moderate production variability, and gradual productivity improvement, demonstrating the potential of AI-based forecasting to support strategic agricultural planning, resource optimization, and sustainable decision-making in the cassava sector.