No abstract

No abstract

The selection of features is used to obtain a subset of features by the removal of irrelevant features with no or less predictive output. Meta-heuristic algorithms are appropriate for the selection of features because feature subset representation is direct and the evaluation is easily accomplished. This paper performed a comparative study on the impact of meta-heuristic optimization algorithms on breast cancer diagnosis using Ant Colony Optimization (ACO) and Particle Swarm Optimization (PSO). The two feature selection algorithms were used to obtain the relevant attributes from the Wisconsin breast cancer (original) dataset. The selected attributes were passed to seven learning algorithms: Support Vector Machine (SVM), Decision Tree (C4.5), Naïve Bayes (NB), K Nearest Neighhood (KNN), Neural Network (NN), Logistic Regression (LR), and Random Forest (RF). The diagnostic model was evaluated based on accuracy, precision, recall, and F1-measure. Experimental showed that the highest accuracy of 97.1388% was obtained in both PSO and ACO using RF classifier, the highest precision value of 0.9720 was recorded in ACO using RF classifier,  the highest recall value of 0.9750 was achieved in PSO using RF classifier, the highest F1-measure value of 0.9700 was obtained in PSO using SVM, the highest kappa statistic of 0.9370 was obtained in both PSO and ACO using RF and the lowest time of 0s was taken to build a model was recorded in PSO using KNN and NB, and also in ACO using KNN. The paper concluded that the breast diagnostic model using PSO and ACO with different learning algorithms revealed that the accuracy of RF outperformed other algorithms. Also, it was shown that ACO produced better precision using RF compared with PSO and PSO gave better recall using RF compared with ACO, PSO recorded an efficient F1-measure using SVM. The best time used to build a model was obtained in PSO for KNN and NB, and ACO with KNN.Keywords— Breast cancer, Data mining, Diagnosis, Feature selection, Meta-heuristic.

No abstract

No abstract