Optimizing ANN Architecture for Classifying Student Stress Levels

Classifying students' stress levels is very important in the world of education. For this reason, this research proposes an Artificial Neural Network (ANN) architectural optimization process to classify students' stress levels by comparing the performance of three optimizers: Ranger, Adam, and Adagrad. Optimization is carried out by exploring variations in hyperparameter values such as learning rate, regularization, number of neurons in hidden layers 1 and 2, and epoch. Model performance testing is carried out using evaluation metrics such as accuracy, precision, recall, F1-Score, and AUC. This research also explores the effects of oversampling using SMOTE-N and the effect of adding hidden layers of up to three layers. The results show that the Adam optimizer with SMOTE-N using two hidden layers produces the best model. Ablation studies were conducted to understand the influence of each hyperparameter, showing that regulation and hidden layers play an important role in avoiding overfitting and improving model performance.