Mining Interestinsg Spatial Patterns From Dynamic Spatial Databases

With the development of GPS and remote sensing technology,a growing number of large spatial databases have been accumulated.Therefore,efficiently utilizing the spatial databases to explore interesting spatial patterns is an urgent task.Spatial data mining is the process of discovering interesting and potentially useful patterns from spatial data.Spatial co-location pattern mining is an important direction of spatial data mining and has been research hot.However,many domains and applications collect their data periodically and continuously,so the spatial databases are changing.Mining interesting spatial knowedge and patterns from dynamic spatial databases is challenging.Firstly,the data between two adjacent time slots will change,so research on method for efficiently incremental mining of co-location patterns is important.However,the changed data will incur changed neighbor relationships and changed prevalence,which brings out the challenge for incremental mining.Secondly,strong symbiotic pattern mining originates from microbiology and is well applied in biology and ecological science.Mining strong symbiotic patterns needs a lot of controlled experiments,observations and domain knowledge.However,symbiotic relationships widyly exist in nature and society.The research on mining strong symbiotic patterns from dynamic spatial databases is important and has not been reported.Moreover,the report on mining competitive pairs and causal rules is mainly based on web data and transactional data respectively,not on spatial data.This dissertation studies mining interesting spatial patterns from the dynamic spatial databases.It includes incremental mining of spatial co-locations,mining strong symbiotic patterns,mining competitive pairs and mining causal rules.The main contributions of this dissertation are summarized as follows:1.We analyzed the course of incremental mining of spatial co-location patterns and proposed new methods on generating candidate patterns,generating changed co-location instances,judging the co-location prevalence and proposing new pruning strategies.We also compared the new methods and known methods with analysis and experiments.Extensive experiments showed the new methods can efficiently and accurately execute incremental mining of spatial co-location patterns.2.We summarized the symbiotic conditions in ecological science or biology,obtained the judging criterion of strong symbiotic patterns,and expressed the judging criterion in dynamic spatial databases.Prevalent co-location patterns show the co-located relationship,and symbiotic relationship is more powerful than co-located.The novel algorithm(basic algorithm)was proposed to mine strong symbiotic patterns from prevalent co-location patterns.For improving the efficiency,an improved algorithm and two pruning strategies were presented.A series of experiments verified the efficiency of proposed algorithms,and compared the strong symbiotic patterns and prevalent spatial co-location patterns on real-world databases.The results showed that the strong symbiotic patterns are more interesting.3.We summarized the basic factors of discovering competitor on web data,which are competitive nature and competitive strength.The idea of mining competitive pairs from dynamic spatial databases was proposed.The formal definition of competitive pair was given and the algorithm for mining competitive pairs was proposed.For improving the mining course,a series of pruning lemmas were proposed.We also compared the competitive pairs and prevalent co-location patterns on real-world databases,and the results showed that competitive pairs are more interesting.4.Causal relationship is the most important,strong and difficult to discover.We mined the causal rules hiding in prevalent spatial co-location patterns from dynamic spatial databases.We proposed the definition of causal rule,and proposed algorithm and pruning strategies to mine the causal rules.Experiments evaluated the performance of algorithm and pruning strategies and compared the causal rules and co-location rules.Finally,we compared prevalent spatial co-location patterns,strong symbiotic patterns,competitive pairs and causal rules on the same real-world database,and got someinteresting results.