Detect irregularly shaped spatio-temporal clusters for decision support

Many real-world applications call for the use of detecting unusual clusters (abnormal phenomena or significant change) from spatio-temporal data for decision support, e.g., in disease surveillance systems and crime monitoring systems. More accurate detection can offer stronger decision support to enable more effective early warning and efficient resource allocation. Many spatial/spatio-temporal clustering approaches have been designed to detect significantly unusual clusters for decision support. In this paper, we focus on more accurately detecting irregularly shaped unusual clusters for point processes and propose a novel approach named EvoGridStatistic. The original problem is mathematically converted to an optimization problem and solved by estimation of distribution algorithm (EDA), which is a powerful global optimization tool. We also propose a prospective spatio-temporal cluster detection approach for surveillance purposes, named EvoGridStatistic-Pro. Experiments verify the effectiveness and efficiency of EvoGridStatistic-Pro over previous approaches. The scalability of our approach is also significantly better than previous ones, which enables EvoGridStatistic-Pro to apply to very large data sets in real-world application systems.