Human-Centered Multi-Sensor Framework for Identifying Driving Patterns Associated with Cognitive Decline Through Quantitative Analysis

Driving places high demands on cognition, making everyday driving behavior a promising domain for detecting Mild Cognitive Impairment (MCI). This pilot proof-of-concept study deployed AutoPi telematics units in 51 older drivers, including 10 with MCI and 41 who were cognitively unimpaired, and monitored them over 28 months. The resulting dataset comprised 20,145 trips captured through GPS, IMU, and OBD-II sensor streams. We developed a multi-stage analytical framework that combined K-means clustering for behavioral profiling, Random Forest for feature ranking, Welch’s t-tests with Benjamini-Hochberg correction, and L1-regularized logistic regression with participant-level leave-one-out cross-validation. The model achieved an AUC of 0.698 (95% CI: 0.493–0.872) and a sensitivity of 0.800. Throttle position variability and mean throttle application emerged as the strongest sensor-derived predictors, each with a Cohen’s d of 0.86, suggesting impaired speed regulation that is consistent with executive dysfunction in MCI. However, the cohort was notably imbalanced by gender, with 9 of the 10 participants in the MCI group being female, indicating that demographic characteristics, especially gender, contributed substantially to the model’s overall discrimination. When gender was excluded, performance declined to an AUC of 0.598, which was nearly identical to the telematics-only result of 0.595. This finding suggests that the driving-behavior signal is meaningful, but modest, once demographic confounding is removed. A cold-start analysis further showed that approximately 50 trips, corresponding to about four months of naturalistic driving, may represent the minimum observation window needed for reliable screening. Subgroup analyses indicated that observed performance disparities were more likely driven by cohort composition than by systematic model bias. Overall, these findings support telematics-based MCI monitoring as a promising direction, while also highlighting the need for validation in larger and more gender-balanced cohorts before clinical deployment.