Abstract

Accurate weather forecasting is crucial for numerous sectors, including agriculture, disaster management, and daily life. This study leverages advanced data mining techniques to analyze and predict weather patterns using extensive historical weather data. Predicting daily weather patterns with high accuracy remains challenging due to the complexity and variability of climate factors. This study aims to identify the most effective machine learning models for this task by comparing various algorithms. Weather data collected over ten years from ten different datasets were analyzed using a diverse set of machine learning models, including rules-based (OneR, Decision Table, JRIP, Ridor), tree-based (J48, LMT, Random Forest, CART), and function-based (MLR, MLP, SVM, LogitBoost, SMO, ANN) approaches. Each model was evaluated based on multiple performance metrics: precision, recall, accuracy, F-measure, True Positive Rate (TPR), False Positive Rate (FPR), True Negative Rate (TNR), and False Negative Rate (FNR). The Random Forest model outperformed others with an accuracy of 92.5%, precision of 91.3%, recall of 90.8%, and an F-measure of 91.0%. The SVM and ANN models also shown strong performance, with accuracies of 90.1% and 89.7%, respectively. Function-based models showed higher robustness in variable conditions, while tree-based models provided better interpretability. Rules-based models, although simpler, yielded lower performance metrics, with OneR achieving the highest among them at 81.2% accuracy.