Compact co-location pattern mining

With the advent of data gathering and analysis, many different domains such as public health, business, transportation, geology and so on are generating large volumes of data. Such large sets of data could contain potentially interesting patterns that can provide useful information to these domains. Many techniques in the area of data mining have been employed to discover useful patterns. Spatial co-location pattern mining has been popularly studied in spatial data mining area. Spatial co-location patterns represent the subset of spatial events whose instances are frequently located together in nearby geographic space. A common framework for mining spatial co-location patterns employs a level-wised Apriori-like search method to discover co-locations and generates redundant information by searching all 2l sub-sets of each length l event set. As the number of event types increases, the search space exponentially increases. This adversely increases the computational time of mining for co-location patterns. In this thesis, we propose two problems for mining compact co-locations which concisely represent co-location patterns. The first problem addresses mining for maximal co-located event sets. The second problem addresses top-k closed mining which finds k closed co-located event sets which have higher prevalence values than other closed co-locations. This thesis develops an algorithm for each problem to efficiently search for the compact patterns. The experiment results show that our algorithms are computationally efficient in finding the compact co-location patterns