<![CDATA[Rheumatoid arthritis (RA) is a chronic autoimmunedisease that requires early detection to prevent permanent jointdamage. The main challenge in predicting RA risk lies in thecomplexity of clinical data, which is multidimensional and oftenimbalanced. This study employs a Hybrid Restricted BoltzmannMachine (H-RBM) approach to improve the accuracy of earlyRA risk prediction. The RBM method is used as a featureextractor to map non-linear relationships within clinical data,which is then combined with three classification algorithms:Support Vector Machine (SVM), Random Forest (RF), andLogistic Regression (LR). The research data was sourced fromKaggle, comprising 4,425 samples that were balanced using theSynthetic Minority Oversampling Technique (SMOTE). Testingwas conducted with three data split ratios (70:30, 80:20, and90:10) using GridSearchCV for hyperparameter tuning. Theexperimental results show that the RBM-SVM model dominatesabsolutely with the highest performance at the 90:10 ratio,yielding an accuracy of 99.82%, an AUC score of 1.0000, and arecall of 100%. The perfect recall rate indicates that the modelis capable of identifying all positive patients without any missedcases (zero false negatives). Meanwhile, the RBM-RF and RBM-LR models achieved optimal performance at an 80:20 ratio withaccuracies of 93.95% and 83.42%, respectively. Overall, RBMintegration proved significant in extracting more informativelatent information compared to raw clinical features, therebymaking a crucial contribution to a precise RA early detectiondecision support system]]>
Copyrights © 2026
