<![CDATA[Weather classification in West Java presents several challenges, particularly related to class imbalance in the dataset and the complexity of meteorological variables. This study aims to improve classification accuracy by proposing a stacking classifier approach that combines Support Vector Machine (SVM) and Random Forest as base learners, with Logistic Regression serving as the meta-classifier. To address the class imbalance, the Synthetic Minority Oversampling Technique (SMOTE) was applied, while model optimization was conducted using GridSearchCV. Weather data from the Indonesian Meteorological, Climatological, and Geophysical Agency (BMKG) for December 2024 was used and processed through transformation, normalization, and outlier handling. The dataset was then split into training and testing sets with ratios of 70:30, 80:20, and 90:10. The stacking classifier without SMOTE achieved the highest accuracy of 86.73%, but suffered from overfitting, indicated by a 13.27% gap between training and validation accuracy. The application of SMOTE improved the recall for minority classes to 76.3% and reduced overfitting, with the accuracy gap narrowing to less than 1%. The most stable performance was achieved with an 80:20 train-test split, where the SMOTE-applied and hyperparameter-optimized model reached an accuracy of 85.97%, an F1-score of 68.99%, and a statistically significant t-test result (p < 0.001). These findings demonstrate that the combination of stacking classifiers, SMOTE, and hyperparameter tuning effectively mitigates class bias and enhances model generalization, outperforming single-model classifiers in handling imbalanced weather data.]]>
Copyrights © 2025
