A Stacking Ensemble Deep Learning Model for the Parkinson Disease Classification

Parkinson's disease (PD) is a prevalent and serious neurological condition. Early signs of the disease have been associated with voice impairments, making the development of diagnostic tools crucial for early detection. Furthermore, intelligent systems designed to classify PD patients and healthy individuals could be valuable for prodromal diagnosis in the future. Individuals with rapid eye movement (REM) sleep behavior disorder (RBD) serve as a good model, as they are highly likely to develop PD. Studies have demonstrated that even minor speech and voice impairments may indicate preclinical PD. In this study, a stacking ensemble deep learning model combined with a Modified Brain Storm Optimization-based feature selection technique is proposed for the effective classification of PD patients. Energy, wavelet Shannon entropy, zero-crossing rate (ZCR), Mel Frequency Cepstral Coefficient (MFCC), and Linear Predictive Coding (LPC) are among the variables that were retrieved. To select features and reduce the number of features, the Modified Brain Storm Optimization method is utilized. Subsequently, a stacking ensemble deep learning model is introduced for classifying PD data. Analysis of the implementation demonstrates the effectiveness and superiority of this novel approach compared to existing classification models, such as ResNet V2, LSTM, Granular Neural Networks, and Elman Neural Networks. In classification tasks, predictive models are used to determine class levels, and this research explores various such models. Combining multiple classification algorithms to create a set of predictive models for each class level is often a more effective strategy. The proposed system shows promise as a highly effective tool for diagnosing PD with exceptional performance.