Classification of User Adherence to Home Hand Rehabilitation Technology Using a Feed-Forward Artificial Neural Network

Mohammad Shams; Daniel K Zondervan; Quentin Sanders

Abstract

Hand impairments resulting from neurological conditions can significantly affect individuals' quality of life. Home-based rehabilitation programs are promising solutions to address these challenges. This study investigated user engagement with MusicGlove, a commercially available wearable grip sensor. We applied deep learning and machine learning techniques to classify users based on their interaction with the device. We categorized users into 'low', 'moderate', and 'power' users and found considerable differences in device usage. For user adherence prediction after one day of device usage, we used a Multi-Layer Perceptron (MLP) deep learning model and traditional machine learning models such as K-Nearest Neighbors (KNN), Support Vector Machines (SVM), and Logistic Regression. The MLP model outperformed other models, achieving an average F1-score of 0.68 in cross validation and a balanced performance on unseen test data with an accuracy of 0.68, precision of 0.66, recall of 0.72, and an F1-score of 0.69 for the 'Low' user class. Our results underscore the need for personalized home-based rehabilitation programs and highlight the effective use of deep learning algorithms in predicting user adherence in home-based digital rehabilitation. This study contributes to the growing body of evidence supporting machine learning applications in healthcare, particularly in patient outcome prediction and treatment personalization.

Keywords: In-Home Rehabilitation; Wearable Sensor; Machine Learning; Patient Adherence

Links
[Full text PDF][Bibtex]