Clinical Intelligence for Parkinson’s Disease with Fully Trained ML Models on Medical Records

Abstract

Parkinson’s Disease is a one of the progressive neurological disorders that often goes undetermined in its early stages due to subtle symptoms, especially in voice and motor control. In this chapter, we explore a practical way to detecting Parkinson’s disorder via supervised-trained Modeling techniques applied to structured biomedical information. The dataset used consists of 569 voices collected from 31 individuals, with each sample described by 22 parameters such as shimmer, pitch frequencies, jitter, noise ratios, and non-linear vocal signal patterns.

These features were pre-processed and standardized before fading into multiple supervised classification models including Logistic Reg, Support Vector classifier, K-Nearest Neighbour classifier, DT, and RF. All trained models were evaluate using real-world performance metrics like accuracy, F1-score, precision, recall, and curve of ROC-AUC. Among all models tested, Random Forest classifier achieved the most reliable results with strong generalization capability on unseen data.

To support practical understanding, model visualizations such as confusion matrices, decision trees, and SVM boundary plots generated. The results show that machine learning models trained on structured clinical data can effectively support early-stage PD detection. This work demonstrates how integrating healthcare records and AI-based predictive tools can assist clinicians in faster and more accurate diagnosis of Parkinson’s Disease, contributing toward more intelligent and accessible medical systems.