An Optimization Algorithm of Synthesizing a Feed-Forward Neural Network to Determine a Human Functional State Using Stabilometry Data

Abstract

Balance is crucial to an individual's quality of life and functional performance. Stability measurement analysis and balance assessment rely on center-of-pressure coordinates and numerical data. Although machine learning algorithms have been applied to analyze stabilization measurements, accurately determining an individual's balance stability remains a challenge despite promising results. This study assesses the efficacy of a classification model—specifically, artificial neural networks (ANNs) utilizing an evolutionary algorithm (EA)—trained on three stability indicators to evaluate human health status. The methodology involved enhancing the learning process of artificial neural networks (ANNs) by dividing the hidden layers into multiple ANNs based on the number of neurons, optimizing them using an evolutionary algorithm, and then combining them to formulate new optimal hidden layers. This method expedited the optimization process and determined optimal designs. This study illustrates that optimal learning phases enhance the selection of appropriate artificial neural network architectures for distinguishing between healthy and diseased conditions, attaining accuracy rates of 99% to 100% for the A-indicator, 98% to 100% for the AW-indicator, and 97% to 100% for the AXY-indicator. The findings demonstrate that the integration of evolutionary algorithms and artificial neural networks markedly enhances predictive accuracy in healthcare, necessitating additional research to corroborate these results.