Baseline Results

We present baseline results for this year’s challenge using physiological signals, specifically Electrodermal Activity (EDA), Blood Volume Pulse (BVP), Respiration (Resp), and Blood Oxygen Saturation (SpO₂). No preprocessing, filtering, or downsampling was applied  to the signals; instead, raw data were used directly for feature extraction based on simple statistical features. 

A Gaussian SVM (RBF kernel) was trained individually on each modality, as well as on a feature-level fusion of all four modalities. The multimodal system was constructed by concatenating features from all modalities prior to classification.  The objective of the baseline system is to distinguish between the three classes defined for this challenge: No Pain, Pain Arm, and Pain Hand. The classification accuracies achieved on the validation and test sets were as follows:  

Set Modality Accuracy

Validation EDA 43.05%

Validation BVP 43.75%

Validation Resp 35.64%

Validation SpO₂ 35.87%

Validation Multimodal 41.66%

Test EDA 39.35%

Test BVP 37.73%

Test Resp 33.56%

Test SpO₂ 35.87%

Test Multimodal 39.81%

Accuracy was calculated as the proportion of correctly predicted samples across the three pain categories (No Pain, Pain Arm, and Pain Hand) out of the total number of validation samples. 

Participants are encouraged to explore different sensor combinations and fusion strategies, as improved performance may be achieved by leveraging complementary information from subsets of the available modalities. 

** Baseline results for the Test set will be provided on the day that the Test set is released. **