CLASSIFICATION OF HUMAN MEMBRANE PROTEIN TYPES USING OPTIMAL LOCAL DISCRIMINANT BASES FEATURE EXTRACTION METHOD

摘要

This paper presents a method of membrane protein feature extraction using a combination of the local discriminant bases (LDB) and three different classifiers. This method has adopted two dissimilarity measures of normalized energy difference and relative entropy to identify a set of orthogonal subspaces in optimal wavelet packets. The energy will be derived from the calculation of the two dissimilarity measures that have overlapping subspaces. This feature, in turn, serves as an input to support vector machine (SVM), decision tree and na?ve Bayes classifiers. The proposed model yields the highest accuracy of 78.6%, 76.25%, 76.72% for dataset S1, S2, and S3 respectively by using SVM. This technique outperformed other feature extraction method for membrane protein type classification for dataset S2 and S3.