Parkinson’s disease prediction model using improved Gaussian process classifier-based optimization framework

To enable timely management and surveillance, Parkinson’s disease (PD) prediction aims to identify early signs of neurodegeneration before the onset of severe clinical symptoms. Machine learning methods can help with accurate diagnosis and progression tracking using imaging, clinical, and sensor information. However, barriers such as the delayed onset of symptoms, patient variability, limited labelled information, and multi-modal complexity hinder prediction efficacy. Longitudinal dynamics, a variety of information formats, and resilience across a range of populations and recording environments are all necessary for successful methods. Hence, this study proposes a Parkinson’s Disease Prediction (PDP) model using an intelligent machine learning framework. The dataset was first gathered from online sources referred to as the Parkinson’s Progression Markers Initiative – Phase 2 (PPMI 2.0). The collected data were pre-processed using digital biomarker resampling and normalization approaches. Segmentation was then performed on the pre-processed data using the multi-atlas label fusion segmentation approach. The features are extracted from the segmented data using graph-based cortical thickness network features. Finally, the PDP model was predicted using a novel Improved Gaussian Process Classifier (IGPC) model. Enzyme Action Optimization (EAO), a nature-inspired optimization algorithm, tunes the parameters of the conventional GPC model. The fitness function behind the entire procedure is considered to minimize the error. The proposed IGPC-EAO for the PDP model is 23.79% and 25.68% better than the other considered traditional models with respect to prediction accuracy and MSE, respectively.