Assessing cognitive load using pupillography frequency features presents a persistent challenge due to the lack of consensus on optimal frequency limits. This study aims to address this challenge by exploring pupillography frequency bands and seeking clarity in defining the most effective ranges for cognitive load assessment. From a controlled experiment involving 21 programmers performing software bug inspection, our study pinpoints the optimal low-frequency (0.06-0.29 Hz) and high-frequency (0.29-0.49 Hz) bands. Correlation analysis yielded a geometric mean of 0.238 compared to Heart Rate Variability features, with individual correlations for low-frequency, high-frequency, and their ratio at 0.279, 0.168, and 0.286, respectively. Extending the study to 51 participants, including a different experiment focusing on mental arithmetic tasks, validated the previous findings and further refined bands, maintaining effectiveness with a geometric mean correlation of 0.236 and surpassing common frequency bands reported in the existing literature. This study represents a pivotal step toward converging and establishing a coherent framework for frequency band definition to be used in pupillography analysis. Furthermore, based on this, it also contributes insights into the importance of more integration and adoption of eye-tracking with pupillography technology into authentic software development contexts for cognitive load assessment at a very fine level of granularity.
Read full abstract