Sensor Data Analysis by means of Clustering

Traffic jams are a big problem of the society nowadays, especially in case of urban traffic. To solve the problem of traffic congestion and air pollution, the intelligent transportation systems (ITS) should be developed and integrated into transport infrastructure. The core element of such ITS is a reliable and accurate forecasting model to predict traffic flow in a short-term period. Lots of historical traffic data can be used as input of the model, in particular daily traffic profiles. Different dates have different traffic flow patterns, and modern prediction models should take into account such temporal variations. This paper investigates the historical traffic flow data, which was obtained from stationary road sensors. The main goal of this research is to obtain more insight into urban traffic by analyzing between day and between month variations in traffic volumes. By means of k-means clustering procedure, we divide daily traffic profiles of each sensor into several groups and examine the obtained clusters with respect to day type (day-of-week, preholiday and holiday) and seasonal variations. For most road sensors, there is a significant difference between the daily traffic flow profiles for working and non-working days. The centroid of the first one demonstrates two prominent flow peaks (morning and evening), whilst the second one represents only one day peak with slow growth and slow decrease of flow rate. We discovered seasonal variation for some road sensors, but it is less pronounced than the variations between weekdays.