Cognitive load detection from wrist-band sensors
In this work, we use machine learning (ML) to detect the cognitive load of a user based on sensor data from a smart wrist-band, sampled during 30 seconds. The data is provided by a challenge at the UbiTtention 2020 workshop of UbiComp 2020; in this paper we describe UW's participation (team Lynx). The defining characteristic of our approach is in the custom features that we extract from the time series. While we do not have any labeled instances for the test users, the fact that we do have multiple time series for each test user, allows us to extract features that measure how much individual time series deviate from the user's average. We combine this extracted information with other time series' features from the literature. We further use feature selection based on Gini impurity and state-of-the-art techniques for training ML models such as Logistic Regression, (Boosted) Decision Trees, Random Forests, and Support Vector Machines, yielding~63% accuracy by 6-fold cross-validation.
Adjunct Proceedings of the 2020 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of the 2020 ACM International Symposium on Wearable Computers
Open Access Status
Li, X., & De Cock, M. (2020). Cognitive load detection from wrist-band sensors. Adjunct Proceedings of the 2020 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of the 2020 ACM International Symposium on Wearable Computers, 456–461. https://doi.org/10.1145/3410530.3414428