Smartphone Sensor-Based Driver and Passenger Classification: Exploring Techniques for Resource-Constrained Devices

Abstract

The utilization of smartphone sensors for transportation safety and efficiency has gained momentum, yet challenges persist in balancing effectiveness with device resource constraints. In this study, we investigate driver versus passenger classification using smartphone sensor data collected in a natural setting, without specific usage instructions. Additionally, we explore methodologies for deploying mobile sensor-based classification systems, considering device limitations. Key tasks involve evaluating small-time-frame windows for data capture, comparing classifier performance with and without feature standardization and selection, and assessing dimensionality reduction techniques. Findings provide insights into optimizing resource efficiency while maintaining classification accuracy in transportation research applications. Specifically, results reveal that shorter data collection periods yield improved classification performance, with notable gains observed as the window size decreases. Furthermore, feature standardization significantly affects classification performance, especially in longer data collection periods, while feature selection consistently enhances accuracy across most experiments. Our findings underscore the importance of tailored data collection strategies and preprocessing techniques for effective smartphone sensor-based classification.