A Detailed Study on Efficiency of ML Based Algorithms in Predicting Parkinson's Disease Using Audio

In this study, we analyse the various machine learning models in order to predictively diagnose the potential of the Parkinson's disease. We use a voice dataset in this study which is used to train the various machine learning models used here. The vocal parameters in the dataset are used to train our models. Our research has given us the following accuracy for each of the given models, Logistic Regression: 89.74%, SVM (Support Vector Machine): 89.74%, Decision Tree: 92.31%. Random Forest, kNN(k Nearest Neighbor) and XGBoost having the same accuracy of 94.87%. Stacking Models: 84.62% and Polynomial Features model: 94.87%. In our study, to analyse and better understand the working and the performance of the models, the metrics used by us for the evaluation of the models are: recall, precision, accuracy and F-1 score. In our research presented in this article, we have come upon the conclusion that the best performing models in the given context are Random Forest along with kNN (k Nearest Neighbor) and XGBoost. To further choose the optimal model among the many, we use a confusion matrix. In our study we have taken into consideration the gravity of the disease and leveraged the use of modern and traditional models to to improve the non invasive methods of predictive diagnosis.