AAI_2025_Capstone_Chronicles_Combined

MENTAL HEALTH RISK DETECTION USING ML​

​

​

​

​

13

training. Finally, the model was trained for 1,000 iterations to support stable convergence and optimize performance across all classes. The TNN is a multi-component neural network that was created to perform well on multi-class classification with a vast amount of features and data. It is trained with 20 epochs and a batch size of 256 to provide sufficient training time, speed, and memory efficiency while avoiding overfitting. The initial dataset was split into 70% training and 30% testing subsets to facilitate performance evaluation. Its core component, the TabNet block, is repeated three times, as illustrated in Figure 4 .

Figure 4 Selected Architecture (TNN) for Mental Health Risk Detection

213