CVAE-Based Hybrid Sampling Data Augmentation Method and Interpretation for Imbalanced Classification of Gout Disease

Gout disease is a highly painful condition worldwide. The use of clinical data for staging gout is crucial in aiding experts with rapid diagnosis and treatment. However, current research has not given sufficient consideration to the accurate staging of gout. On the other hand, imbalanced data presents a challenge in disease diagnosis, which hinders accurate predictive data analysis in real-world medical application. Hence, this study proposes a new predictor model with interpretation for the stages of gout disease (termed PIGD). To address the issue of imbalance, a new hybrid sampling method CBHS that utilises the generative network is proposed. Improved TabNet was used as the foundation for gout detection. The PIGD used the SHAP to propose an interpretable model for the detection of gout, which reveals causal relationships between features to experts and overcomes the limitations of ‘black-box’. The results indicate that proposed PIGD performed the best in the clinical dataset for gout, which get a precision of 86.65%, an AUC of 82.96%, an accuracy of 80%, and a G-means of 77.51%.