A Diagnostic Tool for Rheumatoid Arthritis and Disease Activity Levels using Machine Learning Classifiers

Abstract

Rheumatoid Arthritis (RA) stands out as a severe chronic inflammatory disease potentially leading to disability. It mostly affects the joints in body including small joints of hands and feet such as wrists, finders, and toes and large joints such as knees and shoulders. The symptoms resembles much with other inflammatory arthritis diseases that is a major challenge in its early diagnosis. This paper presents an automated tool for diagnosis of RA and its disease activity levels using machine learning classifiers based on signs and symptoms that RA patients noticed and recorded. The dataset comprises of anonymised clinical data, lab reports, and demographics information of 104 RA patients collected from a local hospital in Multan. This study compares the performance of five supervised machine learning classifers including AdaBoost, Support Vector Machines, Random Forest, Naïve Bayes, and Decision Trees for diagnosis of RA and its activity level. The algorithms selected the clinicaly robust features using the parameter of information gain for classificatio of RA and non-RA patients. Disease activity levels are further categorised using DAS-28 score in terms of severity. The goal of this study is to determine the best suited machine learning algorithm for RA diagnosis that can be used to assist physician in early diagnosis. The performance of AdaBoost classifier was better than other algorithms in terms of precision, recall, F1-score, error rate, and specificity for diagnosis of RA and its various activity levels.