AAI_2025_Capstone_Chronicles_Combined

MENTAL HEALTH RISK DETECTION USING ML​

​

​

​

​

20

The ROC curve, seen in Figure 6 , also reveals that the TNN-Inspired Model handles class imbalance effectively, as evidenced by the minimal spread between class-specific AUCs and the close alignment with the macro-average. Notably, Class 2 (High Risk) displays the highest AUC, which suggests that the model is sensitive to more severe mental health indicators, an important trait for early intervention systems. The smoothness of the curves indicates stable probability calibration across all classes, meaning the model's confidence levels align well with actual outcomes.

Figure 6 TabNet-Inspired Model ROC Curve

220