Weighted feature dimensions according to Fisher's linear discriminant rate and its application on protein sub-cellular localization

The efficiency research about protein sub-cellular localization has become a hot topic recently. Feature extraction plays an important role in the accurate classification or location of proteins. Since the contribution of each feature dimension is different, this paper enlarges the contribution of feature dimensions which have great effect on classification by weighting with its Fisher linear discriminant rate. Then k-nearest neighbor (KNN) algorithm is used to classify the testing sample. The result shows that, compared to direct use of KNN algorithm, KNN with LDA dimensional reduction improves the predicting accuracy rate, and the proposed KNN based on Fisher's linear discriminant rate weighting method with LDA dimensional reduction can further reduce the redundance impact and enhance the accuracy of protein localization.