Diagnosing early parkinson's disease using keystroke dynamics and neuro-fuzzy systems

We present a model for the early diagnosis of Parkinson's disease based on keystroke dynamics and a neural network with weighted fuzzy membership. Noise in the keystroke data was removed using Gaussian filtering, followed by the extraction of relevant features using statistical techniques. The extracted features were then used as inputs to the proposed diagnosis model. Using the initial set of 12 features, the model achieved a sensitivity of 92.8%, an accuracy of 91.9%, and a specificity of 91.0%. To further enhance diagnostic performance, feature selection was performed based on the distance between the center points of the fuzzy memberships. With the optimal subset of eight features, the model achieved a sensitivity of 96.4%, an accuracy of 96.1%, and a specificity of 95.8%, representing improvements of 3.6% in sensitivity, 4.2% in accuracy, and 4.8% in specificity.