Parkinson Disease Detection Based On Speech Using Various Machine Learning Models and Deep Learning Models

Parkinson disease is a neurodegenerative disorder. Parkinson disease is the nervous disorder that affects the movements. Early symptoms of parkinson disease are speech impairment, tremor, bradykinesia, writing changes and so on. Parkinson disease cannot be cured but earlier detection and medication can improve the symptoms. So the early detection of parkinson disease is very important. As the dataset is unbalanced, model can become biased to one class.So balancing of dataset is very important. Dataset can be balanced using oversampling technique. So we propose the prediction of parkinson disease based on speech by using various machine learning with and without oversampling on various dataset and the accuracy and other performance metrics of various models are compared. So the model with better performance metrics is identified. Extreme gradient boosting algorithm acquired the highest accuracy of 96%.Further, the Extreme gradient boosting algorithm is improved by using the various optimization techniques that tunes the hyper parameters of the models. The accuracy obtained after optimization is about 97%.