Feature Extraction and Scalability: Custom VGG Inspired Neural Network for Multiclass Stress Detection

Anxiety, depression, and poor sleep are mental health issues commonly linked to chronic stress. Although HRV (heart rate variability) is widely used to measure stress, HRV measurements alone lack high accuracy. This paper compares two CNN architectures for sequence data classification: a custom 1D-CNN with tuned hyperparameters that achieves 94.23% accuracy and a VGG-16 inspired CNN that improves accuracy to 100%. The study details the methodologies, architectures, and training processes, showing how the VGG-16 model captures finer details in sequence data. CNN models offer advantages over traditional machine learning, including automatic feature selection, less preprocessing, and transfer learning.