Multimodal Machine Learning Framework for Early Detection of Parkinson’s Disease Using Hand Drawings, Vocal Features, ECG, and Gait Analysis

Main Article Content

Ashok R S
Dr. Anil Kumar D

Abstract

This work discusses multi-modal detection of Parkinson's disease through advanced machine learning techniques applied on handwriting, voice, ECG patterns, and gait patterns. Early diagnosis is important since the progression of PD is bound to negatively influence even non-motor functions besides motor functions, thus severely affecting the quality of life. The motor symptoms are further emphasized in handwriting analysis in the form of variations in spirals' drawing angles and the amplitudes of wave drawings, which are compromised of fine control. Among the deep learning models tested, DenseNet-121 was exceptional, with an accuracy of 85.17%, which was more than other architectures such as ResNet, AlexNet, and VGG16. Voice analysis is more on non-motor symptoms, which focus on speech disturbances due to tremors and rigidity. Through different machine learning classifiers such as Support Vector Machines, K-Nearest Neighbors, XGBoost, among others, the slight vocal characteristics related to PD were well-detected. The SVM proved to be the most efficient model with 89.74% accuracy, where both precision and recall were well-balanced, thereby demonstrating its efficiency in this area. The scope was widened beyond handwriting and speech by bringing in ECG data analysis, which could possibly provide further diagnostics. Gradient Boosting, Logistic Regression, and Random Forest were evaluated for the dataset, with Gradient Boosting having shown the highest accuracy at 89.57%. Similarly, with gait analysis through chest-mounted MPU sensor data, new perspectives about motor symptoms arose. In the area of gait analysis, AdaBoost model proved to be better to study gait patterns, showing an accuracy of 85.71%, proving the detection of such minor gait abnormalities that most of the classical assessment tends to miss. Integrating handwriting, voice, ECG, and gait analysis in a study brings a holistic framework that goes beyond clinical approaches in conventional methods of diagnosis. Different use of machine learning models enables the early and accurate detection of Parkinson's disease, making interventions at the right time possible.

Article Details

Section
Articles