Enhancing Agricultural Forecasting: A Hybrid LightGBM- BiLSTM Ensemble for Crop Yield Prediction

When it comes to farmers in Tamil Nadu, it is necessary to predict crop yield correctly since the result of each harvest determines their own livelihood and food provision of the whole area. In an attempt to introduce some form of certainty into this process of high uncertainty, we carried out thorough research to determine the most trustworthy and effective AI-based prediction model of crop yield. We assessed different machine learning models, such as Random Forest, Stacking Ensemble, and hybrid models that stacked gradient boosting and deep learning models. The optimized random forest model had achieved R-squared value of 0.914 which is around 91 percent. More sophisticated approaches, including Stacking Ensembles and hybrid ones, which employed LightGBM to extract features and neural networks to learn about the non-linear data. To achieve a higher level of accuracy, we created a new hybrid system that has LightGBM and a Bidirectional LSTM (BiLSTM) network. The LightGBM element effectively learns the interaction of complex features in agricultural data whereas the BiLSTM element learns the temporal and contextual relationships among the features. An impressive model was obtained with the suggested LightGBM + BiLSTM hybrid ensemble which achieved R-Squared of 0.9214. This study proves that machine learning and deep learning methods can be systematically experimented and integrated to create strong predictive models that can be used to make data-driven decisions about sustainable agriculture in Tamil Nadu and other regions.