Explainable deep learning for Parkinson’s disease detection with DeepNetX2

Parkinson’s disease is a progressive neurodegenerative disease. It has severe effects on motor and speech functions and has an acute need to find precise and interpretable diagnostic tools. In this research, the paper presents a neural network algorithmic representation of Parkinson’s disease detection on the basis of speech biomarkers. The strategy uses the DeepNetX2 model under two setups: a baseline model and an improved methodology, which involves Spearman correlation-based feature selection, normalization, dropout regularization, L2 penalties, and RMSprop optimization. Experimental analyses of the UCI Parkinson speech dataset indicate that the trained DeepNetX2 attains a better performance with 94.74% of accuracy and 96.07% area under the curve, compared with the vanilla implementation and the previous machine learning benchmarks. Explainable artificial intelligence methods made sure that there was transparency and used such tools as SHAP and LIME interpretability, which emphasized important speech features contributing to classification. These findings highlight the promise of streamlined deep learning pipelines with clinical decision support and encourage future studies on multimodal approaches that integrate speech and EEG data.