Space With Pattern Mining Algorithm

The rapid development of computer science, network technology and spatial data collection technique cause the explosion of spatial data which goes beyond people's understanding ability. We are eager to find knowledge from the spatial database and escape from this dilemma so called "rich data and poor knowledge". A brand new study field——spatial data mining comes up due to this request. Spatial data mining refers to the process to extract the implicit knowledge, spatial relationships, or other patterns not explicitly stored in spatial databases. It is a multi-disciplinary and multi-technology comprehensive new cross area which brings together subjects like data mining, machine learning, pattern recognition, spatial database, statistics, artificial intelligence, GIS, remote sensing and decision support systems etc..This thesis first compares and analyzes the study and application situation and development trend of traditional data mining and spatial data mining. As a new study and application field, spatial data minig has brilliant prospect, but its level of development and useage are still not high. Then, this thesis tries to explain the relationship between traditional data mining and spatial data mining by studying their basic principles and methods and draws the conclusion that due to the specialty of spatial data, using the traditional mining techniques directly on spatial data cannot effectively find knowledge. There are two ways to solve this problem:studying new theory, technique and method or tranzactionize the spatial data to make the present methods still work. Taking this as the basic principle, this thesis classifies, analyzes and compares the present spatial co-location data mining algorithms and tries to improve one of them.Co-location data mining is a new important study of spatial data mining. It can be seen as a kind of transformation of traditional association rule mining when applied on spatial dataset. This thesis classifies co-location mining algorithms presented into two:1. Using new interesting measures and new methods to discover co-location patterns; 2. Using maximal cliques to transactionalize the space and combine the traditional frequent item set mining methods to mine co-locations. When studying the algorithm named GlidClique, this thesis finds out its flaws and tries to improve it by using a new algorithm CoreClique. Experiments on synthetic data show the efficiency of this improvement. Based on all the analysis and study, this thesis proposes a new algorithm MCEBOSN (Maximal Clique Enumeration Based on Star Neighborhood) to introduce the maximal clique transactionalization to the co-location mining algorithm which finds out the positive co-location patterns. The analysis of completeness, correctness and time complexity and experimental evaluations on synthetic data show the efficiency of MCEBOSN algorithm. At the end of this thesis the conclusion and the future work directions are given.