Predicting Multiple Parkinson's Disease Tremor Categories Using Advanced Neural Network Models

AbstractParkinson’s Disease (PD) is a progressive neurodegenerative disorder in which tremoris one of the most prevalent and clinically relevant motor symptoms. Early and accuratedetection of different tremor types enables improved disease monitoring and personalizedtreatment. This work presents a Deep-learning–based framework for multi-label classificationof tremor subtypes using the ALAMEDA_PD_tremor_dataset. We implemented threedifferent neural networks namely, 1. Feed Forward Neural Network, 2. Patternnet 3. DeepNeural Network. Multi-label predictions were generated using a one-vs-all strategy, andmodel performance was assessed using Hamming loss, accuracy, confusion matrices, andregression diagnostics. The proposed Deep neural-network model achieved a multi-labelaccuracy of approximately 89%, demonstrating reliable discrimination between tremorclasses despite inter-patient variability and overlapping symptom expression.