Automatic Identification of Use of Public Transportation from Mobile Sensor Data

Abstract

Automatic analysis of a user's activity using data from smartphones has become commonplace. Current methods can distinguish modes of transportation such as standing still, walking, running, and traveling in a motor vehicle. However, there is not yet a way to determine automatically if a person is using public transportation or in a private vehicle. Developing this capability will enable novel ways of promoting public transportation use for sustainability. For instance, this information can be used to provide route-specific product/shopping recommendations/coupons. This work presents a novel means of identifying use of public transportation (bus) using sensor data typically collected using smartphones. The method extracts orientation-invariant features from segments of sensor measurements and then uses a subset of the data to train a random forest classifier. The trained classifier is able to identify which segments of accelerometer and gyroscope data represent instances of public transportation use. This method is evaluated using real-world data collected in the Los Angeles County and Orange County areas in southern California. Our results indicate that the method is able to distinguish the mode of transportation of the user in most cases with an f-score of approximately 96%.