Analysis of adaptive systems based on Driver's workload.

Area of Science:

• Human-Computer Interaction
• Automotive Engineering
• Cognitive Science

Background:

• Adaptive in-vehicle systems aim to enhance driver safety and user experience by dynamically adjusting to workload.
• Workload classification models are crucial for the effective functioning of these adaptive systems.
• Existing research has explored various factors influencing the accuracy of these models.

Purpose of the Study:

• To examine workload classification models and their application in adaptive in-vehicle systems.
• To assess the impact of predictor types, experimental settings, and device types on model accuracy through meta-analysis.
• To propose design guidelines and a framework for workload-based adaptive systems.

Main Methods:

• Conducted a meta-analysis of 31 studies on workload classification models.
• Evaluated predictor types (e.g., physiological data), experimental settings (simulator vs. on-road), and device types (wearable vs. remote).
• Assessed the performance of Random Forest and Neural Network models for binary and multi-class classification.

Main Results:

• Incorporating physiological data significantly improved workload classification model accuracy.
• Random Forest models achieved highest accuracy for binary classification; Neural Networks showed promise for multi-class.
• Adaptive systems using multi-input models effectively adjusted to workload, enhancing safety and user experience.

Conclusions:

• Workload classification models, particularly those using physiological data and multi-input approaches, are effective for adaptive in-vehicle systems.
• Challenges include ensuring model generalizability, addressing system over-reliance, and promoting wider system implementation.
• Future research should focus on developing robust, context-aware systems for real-world driving demands.