Evaluation of Classification Algorithms with Solutions to Multi-Class Imbalance Problems on Academic Dataset using WEKA

This study evaluates the performance of various classification algorithms on academic dataset, while focusing on multi-class imbalance problems. Utilizing the WEKA platform, classifiers are assessed on both original and balanced datasets. The comparison on various algorithms has been performed based on metrics such as accuracy, Precision, Recall, F-measure, and ROC. The findings reveal that balancing the dataset significantly enhances classifier performance. Algorithms like J48, REPTree, JRIP, OneR, and PART consistently achieve high predictive accuracy, with Rule-Based classifiers demonstrating exceptional results. In contrast, classifiers like SVM, NB, and k-NN show lower performance, attributed to limitations such as kernel dependencies and feature correlations. The study concludes that balanced datasets substantially improve classifier reliability and accuracy, highlighting the superiority of Decision Tree and Rule-Based algorithms for educational data mining. These findings pave the way for future research into ensemble-based methods to further enhance classification outcomes.