RELIEF-DISC: An Extended RELIEF Algorithm Using Discretization Approach for Continuous Features

Feature selection is an effective pre-processing technique for dealing with dimensionality reduction. For classification, subset of relevant features must be selected to increase the overall accuracy of classification. RELIEF is considered to be one of the most successful algorithms for assessing the quality of features. RELIEF and the extended algorithms deal with both discrete and continuous features by selecting samples of data from the training set. In this paper, we propose a new algorithm as an extension of RELIEF based on Discretization. Discretization partitions features into finite set of adjacent intervals. Instead of using random sampling for selecting the instance, we suggest to take instance from each interval which reduces the computational complexity and maintains the quality of features. Also, there is no need of user input for sample size parameter. Experimental result shows the performance of the new algorithm in comparison with the existing RELIEF algorithm. The data set is taken from UCI ML repository for experiment.