Pilot Study on Fibromyalgia Disorder Detection via XGBoosted Stacked-Learning with SMOTE-Tomek Data Balancing Approach

Abstract

Emotional distress and functional disability amongst a plethora of other issues have marked themselves as symptoms of fibromyalgia, which is a chronic pain disorder that impacts about 12percent of global population. Characterized by widespread pain, fatigue and sleeping disturbances, fibromyalgia patients use opioids as immediate remedy to the unbearable pain experienced. While, this interplay between metabolomics, stress and pain are quite complex – experts have advised against the continued use of opioids. Thus, the quest for alternate treatment by healthcare experts seeks to use machine learning schemes in identification of fibromyalgia predictors, and monitor patient vitals. We study a stacked-learning scheme that fuse 3-models (Korhonen net, Genetic Algorithm and Random Forest) with XGBoost-regressor. With musculoskeletal dataset retrieved – we fuse Tomek-links with SMOTE for model prediction, which results in Accuracy 0.8002 with F1 0.8091 prior to the utilization of SMOTE; However, model yields perfect Accuracy and F1 with SMOTE-Tomek; And can successfully detect fibromyalgia disorder with enhanced performance/generalization.