Machine Learning to Diagnose Neurodegenerative Multiple Sclerosis Disease

Multiple sclerosis (MS) is a progressive neurodegenerative disease with a wide range of symptoms, making it difficult to diagnose and monitor. Current diagnosis methods are invasive and time-consuming. The use of smartphone monitoring is convenient, non-invasive, and can provide a reliable data source. Our study utilises an open-source dataset, namely-"Floodlight"-that uses smartphones to monitor the daily activities of MS patients. We evaluate whether the Floodlight data can be used in training a machine learning (ML) algorithm for MS diagnosis. After necessary data cleaning, we statistically measured the significance of different tests. Preliminary results show that individual test metrics are helpful for training ML algorithms. Accordingly, we use the selected tests in support vector machine (SVM) and rough set (RS) algorithms. Experimenting with several variations of the ML models, we achieve as high as 69% MS diagnosis accuracy. Since we experiment with SVMs and RSs on individual test metrics, we further report the relative significance of those tests and corresponding ML models suitable for the Floodlight dataset. Our model will serve as a baseline for developing ML-based prognostication tools for MS disease.