Improving Multi-Class Classification with Machine Learning and Metaheuristic Algorithm
Gawalee Phatai, Tidarat Luangrungruang · 2025
This study investigates the performance of various algorithms, including SVM, CSSVM, SVM-TLBO, and CSSVM-TLBO, using ten-fold cross-validation to evaluate their accuracy in predicting complex datasets. The evaluation is based on Mean Squared Error (MSE), Mean Absolute Error (MAE), and accuracy metrics. While both SVM and CSSVM exhibit high accuracy during training (0.9984 and 0.9988, respectively), their performance during testing suffers due to increased error rates, highlighting the issue of local optima. The integration of the metaheuristic Teaching-Learning-Based Optimization (TLBO) significantly improves model performance, reducing error rates and enhancing accuracy during testing. SVM-TLBO and CSSVM-TLBO achieve testing accuracies of 0.9649 and 0.9687, respectively, demonstrating TLBO's effectiveness in overcoming local optima and improving predictions on complex datasets.