<![CDATA[Heart disease is a serious medical issue, and detecting it correctly can make a difference between life and death. Recently, machine learning techniques have begun to be used to help doctors in making a diagnosis, however, there is a problem: medical databases are full of unnecessary, excess, or noisy features. These unnecessary features complicate the prediction of newly arrived patients as they slow down the system and confuse it, which results in over-learning; thus, the system may perform well in predicting the disease based on the data it was trained on but will fail while predicting sick patients. In this paper, we present one of the simplest solutions to that problem. The proposed approach utilizes optimization techniques of multi-objective meta-heuristics for ongoing feature selection integrated with ensemble classification algorithms. In short, the proposed method employs four optimization techniques: Multi-Objective Genetic Algorithm (MOGA), Multi-Objective Grey Wolf Optimizer (MOGWO), Multi-Objective Particle Swarm Optimization (MOPSO), and Multi-Objective Artificial Bee Colony Optimization (MOABC). These algorithms work in parallel to reduce both the number of features selected and the classification error. After acquiring the desired best features subsets, these are forwarded to four different classifiers: K nearest neighbors (KNN), artificial neural networks (ANN), support vector machines (SVM), and naive Bayesian (NB). In the experiments on a standard dataset for heart disease, the best results were reported for the combination of MOPSO with KNN since it achieved an accuracy of 96.66%, precision of 98.24%, and F-measure of 98.15%. The values presented are much better than the ones achieved with the help of other existing approaches. Therefore, the proposed approach is very efficient as it enables elimination of any useless features and focus on the most useful ones for the effective prediction of heart diseases]]>
