Machine Learning for Neural Decoding: Using EEG Signals to Detect Freezing of Gait in Parkinson's Patients

Freezing of gait (FOG) is one of the most serious and poorly understood symptoms of Parkinson's Disease (PD). The main focus of this paper is to examine correlations between electroencephalography (EEG) data and episodes of FOG. Ultimately, the goal of the project is the development of a machine-learning model capable of detecting instances of FOG in Parkinson's patients. EEG data was retrieved from an existing dataset, processed, and analyzed using linear regressions to evaluate relationships between the variables. Then, the data was run through four predictive models: a polynomial regression, a dense neural network, a Bayesian neural network, and an autoencoder. The dense neural network generated the best results for multi-patient data with an accuracy rate of 66%. However, when the Dense neural network analyzed data from an individual patient engaged in one task, its accuracy increased sharply to 80–94%. One potential application of this research would be the development of personal EEG tracking and Parkinson's symptom prediction systems.