A Vocal Features-Based Predictive Model for Parkinson's Disease Diagnosis

This study looks at how audio metrics extracted from speech samples might be used to diagnose Parkinson's disease (PD) using machine learning (ML) techniques. The set contains various acoustic characteristics, including frequency variations, jitter, shimmer and other specific measurements of voice, rather than actual voice recordings. Through rigorous preprocessing and comprehensive exploratory data analysis, the data set was carefully prepared for model training. Many ML methods, including k-nearest neighbour, decision tree, logistic regression, Bayes naves, support vector machines (SVM), have been evaluated and implemented. Ensemble methods, such as voting classification methods, were used to exploit the collective power of several models. It is particularly remarkable that a random forest classifier with the entropy criterion achieved an impressive accuracy rate of about 97.3% on the test set. In addition, LLAMA's integration into the language model interface facilitates real-time prediction and enhanced result interpretation. This integration enables users to rapidly assess the likelihood of PD based on acoustic properties, promoting the development of effective diagnostic tools for clinical and broader applications.