Related Experiment Videos
Explainable machine learning revealing the impact of mental and physical health on arthritis
Md Atik Shams1,2, Sumaiya Fatema2, D M Hasibul Islam2
1Department of Computer Science and Engineering, Notre Dame University Bangladesh, Dhaka, Bangladesh.
Objectives:
To develop a robust and interpretable machine learning framework for arthritis risk prediction and to identify important risk factors associated with the disease.
Methods:
This study used four datasets, which are Behavioral Risk Factor Surveillance System (BRFSS) 2019, BRFSS 2021, National Health Interview Survey (NHIS) 2020 and NHIS 2022. We evaluated 11 machine learning and deep learning architectures including a custom stacked ensemble combined with five resampling techniques to address class imbalance and 9 imputation methods to handle missing data. Area under the receiver operating characteristic curve (AUROC), sensitivity, specificity and balanced accuracy are used to evaluate model performance, with fivefold cross-validation to ensure robustness. SHapley Additive exPlanations (SHAP) provided feature interpretability in prediction.
Results:
The combination of the stacked ensemble, Generative Adversarial Network (GAN) imputation and random undersampling (RUS) achieved the best performance with an AUROC of 0.8007 and sensitivity of 0.8060 in the test set of BRFSS 2019. This combination also showed a balanced performance across both subgroups of male and female by maintaining high performance with AUROC of 0.808 and 0.800, respectively, in the test set of BRFSS 2021. The stacked ensemble model with GAN imputation and either Random Undersampling (RUS) or Random Oversampling (ROS) achieved the best performance across all four datasets. SHAP analysis identified higher age, walking difficulty and high body mass index (BMI) as the top physical factors. Among the psychosocial factors, parental separation showed a high impact on predicting arthritis.
Discussion:
The risk of arthritis is influenced by health, lifestyle and childhood experiences. SHAP shows that difficulty walking, age, BMI and childhood stress play important roles in predicting arthritis risk.
Conclusion:
These results confirm important predictors of arthritis risk and provide a clear, interpretable framework.
