Machine learning-based early detection of Parkinson’s disease using handwriting and vocal features

Abstract

This study investigates handwriting and speech patterns in individuals with Parkinson's disease (PD) using machine learning to enable early disease detection—a critical step for effective treatment. Handwriting analysis centers on motor components, such as spiral angle variation and wave amplitude, which reflect the impaired fine motor control characteristic of PD. Among deep learning models evaluated (ResNet, AlexNet, DenseNet, and VGG16), the DenseNet-121 model achieved the highest accuracy of 85.17% for classifying motor control differences. Voice analysis targets non-motor symptoms, focusing on speech disturbances linked to tremors and muscle rigidity. Machine learning classifiers (SVM, KNN, MLP, XGBoost, Logistic Regressor, and Random Tree) were implemented, with SVM demonstrating the best performance by reaching an accuracy of 89.74% alongside strong precision and recall. Combining handwriting and speech analysis offers a more comprehensive and effective PD diagnosis than conventional clinical approaches, facilitating prompt intervention for improved patient care.