A modified Gray Wolf Optimization algorithm for early detection of Parkinson’s disease

Parkinson's disease (PD) is one of the most common neurodegenerative diseases, causing significant morbidity and mortality worldwide. PD can be diagnosed at an early stage by analyzing patient datasets, such as speech and handwriting samples. In this paper, a modified version of the classical Gray Wolf Optimization (GWO) is proposed with an application to detect early-stage PD through processing such datasets. The new model (MGWO-eP) aims to enhance the algorithm's exploration capability (e) and overcome local optima issues by adjusting a key parameter (P) that controls the search agents’ positions. The MGWO-eP is then applied as a feature selection technique to predict PD in its early stages, using samples of speech and writing. The effectiveness of MGWO-eP is validated by benchmark optimization functions for achieving the global optimum. Then six popular machine learning classifiers are applied to three benchmark PD prediction datasets that include hand-writing and speech samples from people with and without PD, namely HandPD Spiral, HandPD Meander, and SpeechPD. The proposed model achieves best overall accuracies of 96.30% (with voting), 94.45% (with random forest), and 98.31% (with voting), outperforming GWO and particle swarm optimization algorithms as they get stuck with local optimal solutions. The results show that the proposed model is robust and can be used for early detection of PD in patients through analyzing datasets, such as their handwriting and speech to help the patients access treatments early in the disease, prolonging time spent with adequate symptom control and delaying years of disability/morbidity.