Optimized Multimodal Machine Learning Framework for Parkinson’s Disease Detection and Severity Analysis

Parkinson’s disease is a chronic neurological disorder leading to subsequent deterioration of gait, speech and movement. Its early diagnosis is critically important as it can reduce the treatment costs and improve the patient's quality of life. Traditional diagnosis methods depend on clinical observations, which are subjective; hence, possible to overlook initial symptoms of the disease at its early stages. A machine learning based Parkinson’s disease early detection framework is presented in this project by taking a range of biomedical voice features: such as jitter, shimmer, pitch and harmonic-to-noise ratio as input parameters. Parkinson’s dataset from UCI is used for training the system, and the normalization and feature scaling techniques are implemented to enhance the system accuracy. Multiple machine learning algorithms, namely Logistic Regression, SVM, Random Forest classifier, have been compared with each other on the basis of various metrics such as accuracy, Precision, Recall, F1-score and confusion matrix. The results display that SVM and Random Forest Classifier perform well among others. We also deploy our trained framework to a user friendly implementation via a simple web interface for facilitating early diagnosis of the disease using a low-cost, non-invasive and efficient system to benefit the practitioner and the decision maker.