Publication

Abstract : The two most interesting and challenging problems in machine learning that attracts huge attention from industry and academia are attribute selection and classification of imbalanced datasets. Rough set models have gained much of importance in the recent years because they neither use any prior information nor assumptions. Conventional rough set models deal only with discretized data whereas real-world applications use real-valued data. Discretization of real-valued data leads to loss of information that changes the characteristics of the whole dataset. One of the solutions proposed to solve this problem is the concept of tolerance-based rough set. This paper proposes variable tolerance rough set method that computes the tolerance value for each attribute compared to the traditional fixed tolerance-based rough set that uses a fixed value for attribute selection. The class imbalanced dataset is normalized, converted to balanced dataset, and correlation-based filter is used to reduce the dimensionality of the datasets. The proposed method is applied on the dimensionality reduced balanced dataset. The computed statistical measures reveal that the proposed method exhibits better performance compared to fixed tolerance rough set as evident from the experimental results.