An Ensemble Deep Learning Model Using CNN, LSTM and Autoencoder for EEG-Based Parkinson’s Disease Classification

Parkinson's disease is a neurological disorder hat effects the movements including shaking, stiffness, difficulty while walking and speaking. This condition will occur when the nerve cells in the brain start breakdown which results in loss dopamine. The diagnosis of Parkinson's disease using RI is one of the costliest methods, there is an alternative method which is cost efficient called as EEG Electroencephalography). Using EEG will helpful in in detail study of brain using brain waves. Resting state EEG has shown potential cost-effective method for diagnosing Parkinson's disease. Here we constructed a Ensemble learning for diagnosing Parkinson's disease. By the help of soft voting the model will decide whether the person is affected by PD or not. e conduct experiment on publicly available datasets which consists of 71 PD patients and 37 healthy control persons. We achieved an accuracy on 98.6%, precision of 98.8% and F1-core of 0.9864. Unlike conventional ensembles, our model introduces a weighted soft voting mechanism that emphasizes Autoencoder-derived latent features, which offer robust, denoised representations of EEG signals. While CNN captures spatial patterns and LSTM models temporal dependencies, the Autoencoder uniquely enhances generalization by compressing and filtering noisy EEG data. To the best of our knowledge, this s the first hybrid model to integrate CNN, LSTM, and Autoencoder for EEG-based Parkinson’s diagnosis, demonstrating superior performance validated through ablation and comparative analysis.