Baseline Results
We present baseline results for fNIRS, video, and their combination. For fNIRS data, only oxyhemoglobin (HbO) was utilized in the analysis. A Gaussian SVM (rbf) was trained, yielding an overall classification accuracy of 43.2% for the fNIRS-only baseline system.
For the video data, features were extracted using the Py-Feat facial expression analysis toolbox. Subsequently, a Gaussian SVM classifier was trained, resulting in an overall classification accuracy of 40.0% for the video-only baseline system.
A feature-level fusion approach was employed, wherein fNIRS and video features were concatenated. Following this, a Gaussian SVM classifier was trained, leading to an overall classification accuracy of 40.2% for the multimodal system.
The following table outlines the baseline performance on the validation set for fNIRS-only, video-only, and multimodal (fNIRS & Video) feature fusion systems. The overall classification accuracy represents the proportion of all samples correctly classified (No_Pain, Low_Pain, High_Pain) divided by the total number of samples in the set.
Set Modality Baseline
Validation fNIRS 43.2%
Validation Video 40.0%
Validation Multimodal 40.2%
Test fNIRS 43.3%
Test Video 40.1%
Test Multimodal 41.7%
Baseline results for the Test set will be provided on the day that the Test set is released.