OPTIMIZING SUPPORT VECTOR MACHINE FOR IMBALANCED DATASETS BY COMBINING POSTERIOR PROBABILITY AND CORRELATION METHODS

The challenge of classifying imbalanced data persists in machine learning, particularly in critical applications such as medical diagnosis, fraud detection, and anomaly identification, where detecting the minority class is essential. Conventional classifiers like Support Vector Machine (SVM) tend to favor the majority class, leading to reduced sensitivity in identifying minority instances. This study introduces Posterior Probability and Correlation-Support Vector Machine (PC-SVM), a novel approach that integrates posterior probability estimation with correlation analysis to enhance SVM’s performance on imbalanced datasets. Unlike traditional SVM models, which struggle with class imbalance and require additional data balancing techniques, PC-SVM dynamically adjusts classification thresholds using posterior probability values and correlation-weighted features, simplifying the classification process while improving its effectiveness. The effectiveness of PC-SVM was evaluated using multiple imbalanced datasets from KEEL, UCI, and Kaggle repositories. Results demonstrate that PC-SVM achieves 100% recall for the minority class, significantly outperforming traditional SVM, which attained only 80% recall on average. This 20% improvement in recall underscores PC-SVM’s ability to mitigate the imbalance issue without relying on oversampling or cost-sensitive adjustments. Furthermore, PC-SVM exhibits consistent performance across various evaluation metrics, including accuracy, precision, recall, and F1-score, ensuring robust classification results. By improving the detection of minority classes, PC-SVM offers a transformative solution for real-world applications that demand high sensitivity in identifying rare but crucial instances. Its ability to maintain classification integrity without additional balancing techniques positions it as a valuable model for industries such as healthcare, finance, and cybersecurity, where accurate minority class recognition is critical.