Finding Meaningful Robust Chunks from Driving Behavior Based on Double Articulation Analyzer

The double articulation analyzer is a machine learning algorithm which can extract double articulation structure from time series data based on nonparametric Bayesian approach. The method was proved to detect intentional changes of a driver from time series data recorded by an instrumented vehicle. In this paper, we segment time series data obtained during a driver drove a car through two types of courses using the double articulation analyzer, and analyze the extracted robust chunks by comparing with tags which were added to the recorded data by human participants.