Adaptive Technique for Feature Selection in Modified Graph Clustering-Based Ant Colony Optimization

Abstract

Modified graph clustering ant colony optimization (MGCACO) algorithm is an unsupervised feature selection (UFS) algorithm used in determining a subset of effective genes from microarray data. The feature subset construction is based on the ant colony optimization (ACO) algorithm, which guides the search process from clusters. However, the MGCACO algorithm is unable to choose all significant features from the clusters to form an optimal feature subset. This paper proposes an enhanced graph clustering ACO (EGCACO) to overcome the problem of feature selection in the MGCACO algorithm. A principal point of this algorithm is utilizing an adaptive selection technique that guides ACO for subset construction from the clusters of features. The adaptive technique for ant selection is based on the state of the search space. Experimental results indicated that the proposed EGCACO achieves the highest classification accuracy than five other common UFS algorithms on four classifiers, where it obtained 87.13%, 86.19 %, 87.38 % and 90.80 % for support vector machine, k-nearest neighbor, decision tree and random forest classifiers, respectively. In particular, the proposed algorithm can select the genes of the deoxyribonucleic acid microarray with consideration of relevance and redundancy among the genes. Therefore, the proposed EGCACO can be implemented to handle the high dimension feature space, such as image processing, text classification, and microarray data processing, which is critical for good and reliable results.