A Survey on Clustering Techniques for Situation Awareness

Situation awareness (SAW) systems aim at supporting assessment of critical situations as, e.g., needed in traffic control centers, in order to reduce the massive information overload. When assessing situations in such control centers, SAW systems have to cope with a large number of heterogeneous but interrelated real-world objects stemming from various sources, which evolve over time and space. These specific requirements harden the selection of adequate data mining techniques, such as clustering, complementing situation assessment through a data-driven approach by facilitating configuration of the critical situations to be monitored. Thus, this paper aims at presenting a survey on clustering approaches suitable for SAW systems. As a prerequisite for a systematic comparison, criteria are derived reflecting the specific requirements of SAW systems and clustering techniques. These criteria are employed in order to evaluate a carefully selected set of clustering approaches, summarizing the approaches' strengths and shortcomings.