Improving effort estimation of Fuzzy Analogy using feature subset selection

Abstract

Feature selection has been recently used in the area of software development effort estimation for improving the accuracy and robustness of prediction techniques. The idea behind selecting the most informative subset of features from a pool of available effort drivers stems from the hypothesis that reducing the dimensionality of datasets may significantly minimize the complexity and time required to reach to an optimal and accurate estimation. This paper compares two relatively popular feature selection techniques (Forward Subset Selection and Backward Feature Elimination) used with Fuzzy Analogy for software effort estimation. This empirical comparison is done over eight well-known datasets with the Jackknife evaluation method. The results suggest that Fuzzy Analogy using feature subset selection generates more accurate estimates in terms of the Standardized Accuracy (SA) and Pred(p) criteria than Fuzzy Analogy without using feature subset selection regardless of the data set used. Moreover, this study found that the use of Forward Feature Selection, rather than Backward Feature Elimination, may improve the prediction accuracy of Fuzzy Analogy and reduce the number of features selected.