Similarity-based classifier combination for decision making

This study focuses on combination schemes of multiple classifiers to achieve better classification performance than that obtained by individual models, for real-world applications such as toxicity prediction of chemical compounds. The classifiers studied include kNN (k-nearest neighbors), wkNN (weighted kNN), WNNModel (kNN model-based classifier), and CPC (contextual probability-based classifier), which are all similarity-based methods. We firstly review these learning methods and the methods for combining the classifiers, and then present three similarity-based combination methods as the basis of our experiments. The experimental results have shown the promise of this approach.