HP-SEFB: hyper parameterized stacking ensemble with fuzzy-Boruta for rheumatoid arthritis classification

Abstract

Rheumatoid Arthritis (RA) is an inflammation and auto-immune chronic condition affecting the entire body, mostly joints. If untreated, it can lead to joint pain and bone erosion. The main objective of this work is to develop a classification model that uses effective feature selection and hyperparameter optimization to accurately predict early RA progression based on disease activity scores. The innovative Fuzzy-Boruta strategy combines fuzzification with the Boruta algorithm to improve feature selection. A novel Hyper Parameterized Stacking Ensemble with Fuzzy-Boruta (HP-SEFB) feature engineering that combines Gaussian Naive Bayes, Support Vector Classifier and Decision Tree as base learners, with Logistic Regression serving as the meta-learner is proposed. HP-SEFB performs better when compared with the stacking ensemble with a score of 95.9% for accuracy, 95.9% for precision, 95.91% for recall and 95.9% for F1 score. Results are validated using the 10-fold cross-validation. The Area Under the Curve values for each class are 99.3% for Remission, 98% for Low Disease Activity, 99% for Medium Disease Activity and 99.6% for high disease activity. The model gives a 94.5% Matthews Correlation Coefficient in predicting true and false predictions across all classes. Also, Cohen's Kappa Score of 94.5 % shows the model’s ability in multiclass classification.

Keywords

• Feature engineering
• hyperparameter optimization
• stacking ensemble
• rheumatoid arthritis