Extracting information from driving data using k-means clustering

There is an increasing availability of data, but for making decisions and other tasks we need information. Hence, we require to analyze the data and extract parts or come up with relations between different pieces. In this paper, we focus on information extraction within the automotive industry. In particular, we report on applying k-means clustering for identifying episodes in vehicle data. An episode is considered to be a time interval where a vehicle is performing an activity worth being distinguished. The underlying idea is to cluster the data such that we are able to extract such similar situations like breaking before a crossing only considering vehicle data. We discuss a method that allows extracting such episodes capturing actuator and sensor readings over time. Besides introducing the underlying method, we present obtained empirical results making use of a freely available dataset showing that the extracted episodes have indeed a meaningful interpretation.