Enhancing Protein Fold Prediction Accuracy Using an Ensemble of Different Classifiers

Protein fold prediction problem is considered as a key point to protein structure recognition and structural discoveries. Recent advances in pattern recognition field brought a great interest to apply pattern classification techniques to tackle this problem. From the pattern recognition point of view, the protein fold prediction problem can be expressed as a multi-class classification task that can be solved by using feature selection, feature extraction and classification approaches. In this paper, a new classifier ensemble, based on combination of five different classifiers, namely: Naïve Bayes, Multi Layer Perceptron (MLP), Support Vector Machine (SVM), LogitBoost, and AdaBoost.M1 combined with five combinational policies (Average of Probabilities, Product of Probabilities, Minimum of Probabilities, Maximum of Probabilities and Majority Voting), is proposed to tackle this problem. To study the effectiveness of the proposed method and compare our results with previously reported results, the dataset produced by Ding and Dubchak is used. Experimental results show that the proposed method (using majority voting as its combinational policy) enhances the protein fold prediction accuracy better than most of the other classification methods found in the literature, using the same set of features (employed by Dubchak et al.).