Development of an optimized Bernoulli Naïve Bayes classifier for threshold-based false onset rainfall prediction

This study presents the development of an optimized Bernoulli Naive Bayes classifier for predicting threshold-based false onset rainfall, a phenomenon critical to farming. The methodology followed in this research includes data collection, preprocessing, feature selection and threshold analysis, model development, model optimization, and evaluation.The primary focus of this research was the optimization of this model to improve its performance. Leave-one-out cross-validation was employed to systematically validate the model by training it on all but one instance and testing it on the excluded instance, ensuring robust performance evaluation. Grid search was used for hyper parameter tuning to identify the optimal parameters that maximize model accuracy. Alpha smoothing was applied to handle zero probabilities, ensuring the model's generalization to unseen data. The model was evaluated using key performance metrics, such as accuracy, precision, recall, and F1 score. Experimental results revealed that the optimized model achieved significant improvements in predictive accuracy and reliability over baseline implementations. This optimization framework highlights the model's computational efficiency and its suitability for real-time applications. The findings establish the potential of the optimized model as a powerful tool for addressing challenges associated with false onset rainfall prediction. Unlike deterministic models, this research emphasized probabilistic reasoning, introducing a novel approach to rainfall prediction.