Artificial Neural Network training using metaheuristics for medical data classification: An experimental study

Abstract

The Artificial Neural Network (ANN) is an important machine learning tool used in medical data classification for disease diagnosis. The learning algorithm in ANN training plays a crucial role in classification performance. Various approaches have been successfully applied as a learning algorithm for ANN training. This paper performs an experimental study that investigates the performance of different metaheuristics as learning algorithms to train the ANN for medical data classification tasks. The experiments are carried out on 15 well-known medical datasets. A comparative study is conducted with the classical Levenberg–Marquardt (LM) and other thirteen recent and relevant metaheuristics. Different evaluation criteria such as accuracy, sensitivity, specificity, precision, Geometric Mean, F-Measure, false-positive rate (FPR) are considered for performance estimation. The classification results are analyzed using Multi-Criteria Decision Making (MCDM) method, and the results with analysis establish that the Equilibrium Optimizer algorithm outperforms all the other algorithms included in the comparative study.