RICE YIELD PREDICTION USING LSTM AND ADABOOST

Accurate crop yield prediction is vital for global food security and sustainable agriculture, yet traditional methods struggle with the complex, non-linear nature of agricultural systems. This study employs Long Short-Term Memory (LSTM) networks and Adaptive Boosting (AdaBoost) to enhance rice yield prediction using soil data, offering improved alternatives to conventional models. LSTM captures temporal dependencies in sequential data, such as climate and soil conditions, while AdaBoost’s ensemble method iteratively improves predictive accuracy by focusing on misclassified instances. The models were trained and tested on a dataset consisting of climatic variables, soil properties, and rice yield data collected over a decade. Their performance was evaluated using metrics such as Root Mean Squared Error (RMSE) and R². The LSTM model achieved an R² of 0.520 and RMSE of 0.126, outperforming AdaBoost's R² of 0.415 and RMSE of 0.139. These results highlight the effectiveness of LSTM in predicting rice yields, especially for capturing time-dependent variables. This research contributes to improving data-driven agricultural practices by offering reliable predictions that can inform better decisionmaking in resource management and farming strategies, ultimately supporting global food systems.