A Novel Stress-Level-Specific Feature Ensemble for Drivers’ Stress Level Recognition

This paper proposes a novel feature set for drivers’ stress level recognition. The proposed feature set consists of data-independent and almost uncorrelated feature pairs for each stress level with very strong intra-class and relatively weak inter-class correlations, constructed by realizing a correlation analysis on the popular features studied in the literature. By using the proposed feature set, a maximum of 100% stress level recognition accuracy is achieved with an average increment of 24.85% while a mean reduction rate of 88.01% is satisfied in false positive rate compared to the full feature set. These outcomes clearly show that the proposed feature set can confidently be integrated into the driving assistance systems