Abstract:
Designing a solution for an optimization problem requires two main aspects; the optimization technique (e.g., search strategy) and the evaluation criteria (i.e., objective function). In this paper, an enhanced binary version of a recent metaheuristic algorithm, the Harris Hawk Optimization algorithm (EBHHO), is presented to find a (near) optimal solution for the Feature Selection (FS) problem. Moreover, three different classifiers called K-nearest neighbors (kNN), Decision Trees (DT), and Linear Discriminant Analysis (LDA) were used as evaluation criteria to formulate the objective function. In addition to reducing the dimensionality of the dataset using the FS technique, the Adaptive Synthetic (ADASYN) oversampling technique was used to enhance the quality of the learning algorithm by re-balancing the dataset. A set of well-known datasets in the field of Software Fault Prediction (SFP) were used to validate the efficiency of the proposed approach. The obtained results showed that EBHHO is superior over the basic HHO as well as proved the ability of the EBHHO algorithm to produce the best result among a set of well-known optimization methods.