This repository presents the implementation of the paper titled "Modular Deep Residual Network for Driver Stress Detection based on Photoplethysmography" We assessed our model's performance on two distinct publicly available datasets (AffectiveROAD and WESAD), achieving accuracies exceeding 98% for both, surpassing existing literature. The emphasis was on utilizing only the PPG signal. Our model incorporates Multi-branch Residual Blocks, where each branch employs convolution layers with varying filter sizes, enabling the model to capture a more diverse range of data.
For detailed results and a comparison with current literature, refer to the following spreadsheet: https://docs.google.com/spreadsheets/d/1suvkRUzqYLREeHeM2TfSO7g9sYqTzjjEeeH3HNNC4i8/edit?usp=sharing