Modelling of Accelerometer Data for Travel Mode Detection by Hierarchical Application of Binomial Logistic Regression

Household trip data collection is essential for design and construction of transportation infrastructure. Conventionally, this information is collected by travel surveys, which require the respondents to answer a list of questions targeting their daily travelling. As the responses depend on the memory of the respondents, inaccuracies usually occur during the reporting process. To improve the accuracy of the collected data, a lot of research is currently being focused on inferring the important information from data collected automatically with the help of devices like smartphones. The current study proposes a new method for identifying the travel mode, by applying the binomial logistic regression in a hierarchical manner, using the data collected by the accelerometer of the smartphone. Three methods of application are discussed, namely ranking, one against rest and one against all. Apart from train, all the other modes are successfully modelled with goodness of fit approaching to 1. Low goodness of fit in case of train is due to the wide range of accelerations recorded. Although, all the three methods exhibit good outcomes, one against all method provides relatively better results.