An Effective Combination of Multiple Classifiers for Toxicity Prediction

This paper presents an investigation into the combination of different classifiers for toxicity prediction. These classification methods involved in generating classifiers for combination are chosen in terms of their representability and diversity which include the Instance-based Learning algorithm (IBL), Decision Tree learning algorithm (DT), Repeated Incremental Pruning to Produce Error Reduction (RIPPER), Multi-Layer Perceptrons (MLPs) and Support Vector Machine (SVM). An effective approach of combining the different classifiers using Dempster’s rule of combination has been proposed and evaluated on seven toxicity data sets came from real-world applications. The experimental results show that the performance of the combination of the different classifiers on seven data sets can achieve 59.95% classification accuracy on average, which is 2.97% better than that of the best classifier generated by SVM, among five classification methods studied.