An ML-Based Expert System for Predicting Rainfall

The accurate measurement of rainfall becomes essential to conduct effective water resource management as well as to prepare and plan for agricultural activities and disasters worldwide. The difficulty for traditional forecasting systems lies in comprehending the complex non-linear relationships that exist in worldwide rainfall patterns. This research evaluates the use of different Machine Learning (ML) techniques to optimize the accuracy after the prediction of worldwide rainfall measurements. The dataset, obtained from Kaggle, comprises diverse weather parameters such as wind_kph, precip_mm, cloud, uv_index, etc. Comprehensive preprocessing, including data cleaning, feature selection, normalization, and categorical encoding, was performed to enhance data quality and model efficiency. The dataset was split into training and testing sets using three different ratios: 70:30, 80:20, and 90:10. Six ML algorithms were implemented and evaluated: Logistic Regression (LoR), Decision Tree (DT), Extra Trees (ETs), MLP Classifier (MLP), and XGBoost. Model performance was assessed using 10-fold cross-validation mean accuracy and test accuracy. Among the models, XGBoost achieved the highest test accuracy of 91.43% in the 90:10 split after hyperparameter tuning. The comparative analysis demonstrates that certain models deliver more reliable rainfall estimation when relevant features are carefully selected and optimized.