Feature selection using stochastic diffusion search
Haya Abdullah Alhakbani, Mohammad Majid al‐Rifaie · Proceedings of the Genetic and Evolutionary Computation Conference · 2017
The method introduced in this paper uses stochastic diffusion search (SDS) to select the most relevant feature subset for the classification task. In this algorithm, SDS is adapted to find a suitable feature subset. Moreover, support vector machine (SVM) is used as a classifier to evaluate the predictive accuracy of the agent. The proposed method exhibits a statistically significant outperformance when compared with the performance of the classifier without the SDS-powered features selections. Additionally, the results have been also compared with other methods from the literature over nine datasets. It is shown that the proposed SDS based feature selection (SDS-FS) offers a competitive performance with other methods on datasets with feature size greater than 10. The behaviour of the proposed algorithm has been investigated in the context of global exploration and local exploitation.