An Improved Instance Based K-Nearest Neighbor (IIBK) Classification of Imbalanced Datasets with Enhanced Preprocessing

摘要

The presence of data with skewed class distributions is a problem common to a variety of fields, including Bioinformatics, Computer science, Text classification, Remote-sensing, and Manufacturing industries. In Bioinformatics applications, the numbers of non-interacting proteins (majority class) are greater than number of interacting proteins (minority class) in predicting the protein-protein interactions. These results to imbalance dataset problem. Standard classification learning algorithms are often biased towards the majority class and therefore there is a higher misclassification rate for the minority class instances. In this work we focused on the minority class, for solving the imbalanced dataset problem. This research work consists of two stages-preprocessing and classification of imbalanced datasets. The imbalanced datasets were preprocessed using the SMOTE technique and it has been improved to show better results. Then the resultant dataset is classified using the existing J48 decision tree, Instance Based K-nearest neighbor (IBK), Support Vector Machine (SVM) and the proposed (IIBK) algorithm. Four datasets have been used for evaluation of the proposed algorithms and our results outperformed the other methods under the statistical comparative analysis.