Machine learning models to forecast cotton yield for Punjab

Abstract

Accurate and early predictions in agriculture are essential for sustainable farming and optimizing field management. Crop yield prediction significantly impacts the farmer’s decisions on crop insurance, storage demand and other important factors during the growing season. Due to non-linearity in crop yield, the use of non-linear models for forecasting purposes has become popular these days. In this paper, ANN and LSTM models were trained using weather parameters to forecast the cotton yield for Punjab, India. Predicting the yield with minimum error is a main challenge. ANN (10 10 10 1) model with ReLU activation function in hidden layers performed better than other forecasting models with a minimum MSE (0.0182). The analysis using the NN model concluded that the weather parameters played an important role in affecting the plant growth. These variables may enhance or reduce the yield significantly. Sensitivity analysis showed that relative humidity was the most important weather parameter followed by rainfall.