A hybrid Grey Wolf Optimization and Particle Swarm Optimization with C4.5 approach for prediction of Rheumatoid Arthritis

Abstract Rheumatoid Arthritis (RA) is a type of dreadful autoimmune disease that affects the entire human body, especially joints. Early diagnosis of RA is a challenging task for General Physicians, since the actual triggering mechanism is unpredictable. The capability of C4.5 was explored using the hybridization of Grey Wolf Optimization (GWO) - Particle Swarm Optimization (PSO) to develop an effective RA prediction system. In this work, firstly, PSO was applied for selecting the diversified initial positions. Secondly GWO was used to update the current positions of the population from the search space to get the optimal features for better classification. Subsequently, the selected features were given as an input to the C4.5 approach and developed a final RA predictor model. The proposed HGWO-C4.5 was meticulously examined based on real time patient’s data, which included factors that influence RA prediction by utilizing both RA and Non-RA information. To validate the proposed HGWO-C4.5, with other meta-heuristics based methods including GWO based C4.5, PSO based C4.5 and individual C4.5 method. The Cross-validation results show that HGWO-C4.5 has achieved an overall average accuracy of 86.36% from three other approaches, which was ∼ 6%–14% higher than those attainable using the individual predictors. Furthermore, HGWO-C4.5 has achieved an overall average accuracy of 84% on independent datasets evaluation, which was 8.61% higher than those yielded by the state-of-the-art predictors. This is the first predictor model that includes feature selection and classification for RA prediction to the best of our knowledge.