Differential evolution and ACO based global optimal feature selection with fuzzy rough set for cancer data classification

In the biomedical research field, feature selection plays the predominant role in prediction of diseases. The main objective of this paper is to predict cancer from microarray gene expression data by proposing two feature selection algorithms, namely (1) differential evolution with fuzzy rough set feature selection and (2) ant colony optimization with fuzzy rough set feature selection algorithms, which solve the multi-objective optimization problems. The first algorithm represents the hybridization of differential evolution and fuzzy rough set and aims to select the global optimal features by applying the fuzzy rough evaluation function as the fitness function. The second algorithm, i.e., hybridization of ant colony optimization and fuzzy rough set, selects global optimal features by applying the fuzzy rough evaluation function as the fitness function. The performance of proposed two features selection algorithms is evaluated with various classification metrics, which are computed from decision tree classifier using tenfold cross-validation. Five datasets are applied to analyze the performance of the feature selection algorithms. The datasets used are diffuse large B cell lymphoma, breast cancer, Leukemia and small round blue-cell tumors which are cancer datasets. In addition, a non-medical dataset, namely Gisette, is also used to demonstrate the generalization capability of the proposed algorithms. The metrics used for comparison are, namely, accuracy, precision, recall, f-measure, specificity, processing time and receiver operating characteristics. The performance comparison evidenced improved performances for the proposed algorithms. Similar to the hybridization of differential evolution and ant colony optimization with fuzzy rough set, particle swarm optimization can be extended in the future.