Evaluating statistical and machine learning models for paddy yield forecasting in Kerala

Abstract

Paddy is the most important food crop in Kerala which plays a significant role in the food security of the state. The present study analyses paddy yield data from 1956-57 to 2022-23, along with meteorological variables such as rainfall and maximum temperature, to develop a forecasting model for paddy yield in Kerala. The study employed statistical and machine learning approaches for the model development. Various time series models have been developed using the ARIMA, ARIMAX, NNAR, and NNARX methods. The selected models are compared for the relative performance using the metrics like RMSE and MAPE. All the models have exhibited good performance in the model building phase, with minimal RMSE and MAPE values. The ARIMA (0,1,1) model with a constant and the ARIMAX (0,1,1) model with a constant exhibited comparable accuracy and model fit in both the training and testing phases. Considering both accuracy and simplicity, the ARIMA (0,1,1) model has been identified as the optimal model for forecasting paddy yield in Kerala.