Efficient Weather Forecasting using Machine Learning Models Using Data Balancing

Precise weather prediction is a crucial study area for both disaster management and the tourism industry. The effectiveness of forecasting relies on various factors considered during the prediction process. This study concentrates on employing classification techniques to predict rainfall and associated categories. Due to the imbalanced nature of most raw data, many models struggle with prediction accuracy. To address this, the research proposes utilizing data preprocessing and the SMOTE oversampling method to balance the data and expand features. The study aims to evaluate the performance of several Machine Learning (ML) classifiers for weather forecasting. Five weather categories are used: rain, drizzle, snow, sun, and fog. The xGboost and Random Forest classifiers demonstrate the best results. The study also compares the performance of ANN, SVM, and decision tree-based classifiers. Accuracy, precision, recall, and F1 score serve as performance metrics. The proposed xGBoost approach achieves a maximum accuracy of 95%, while RF and ANN-based predictions show a 3-4% improvement.