An Effective Search Economics Based Algorithm for Feature Selection

The basic idea of feature selection, as the name suggests, is to select an appropriate subset of features of a problem in question so as to reduce the data complexity, thus accelerating the analysis process. Being an NP-hard problem, it has become a critical research topic in data mining for years. In this study, we propose a novel, effective metaheuristic algorithm, called search economics for feature selection, for solving this problem. The proposed algorithm is built on the idea of investing in different sectors of the market based on the potential of each sector; in other words, it is built on the idea of searching different regions in the solution space based on the so-called potential of each region. By using such a search strategy, the proposed algorithm is less likely to fall into local optimum at early iterations while at the same time retaining the search diversity during the convergence process. The simulation results show that the proposed algorithm provides a better result than all the other state-of-the-art feature selection algorithms compared in this study in solving the eighteen well-known UCI datasets.