Mining High Quality Spatial Co-location Patterns

In recent years,the rapid development of mobile devices,wireless sensors,remote sensing mappers,global positioning systems and geographic information systems,are leading to an increasingly tremendous storage of spatial data.Spatial data mining and knowledge discovery aims to extract implicit,non-trivial and interesting patterns and features from spatial database.Spatial co-location pattern represents a subset of spatial features whose instances are frequently located together in geographic neighborhoods.As an important spatial data mining task,spatial co-location pattern mining technique aims to reveal previously implicit but useful relationship among spatial features and used in extensive domains such as ecology,mobile commerce,public safety,decision making,public service urban planning,public health,and spatial decision making,etc.Due to the generality and diversity of application scenarios,many researchers begin to focus on the availability of co-location pattern mining results in different scenarios.Most previous studies take the prevalence of co-locations as sole interestingness measure,which leads a large number of mining results who can only indicate the spatial co-existence relationship of features in co-location patterns.In practical applications,such mining results which are less targeted and less Interpretable can hardly guide the user's actions.Thus,in order to acquire co-location patterns with high quality which are more interpretative and pertinent,it is necessary to consider the interaction in instances,features and co-location patterns,extract the interesting association between features.In this thesis,by respectively taking into account the interaction between features in a co-location patterns,the interaction of the co-location instances and the other instances in their neighborhoods,and the interaction in different co-location patterns,we propose three methods,to discover the co-location patterns with dominant features,dominant co-location pattern and combined co-location patterns.Thus,we can discover a kind of high-quality co-location patterns which are concise,targeted,Interpretable.The main research contents and contributions in this thesis are summarized as follows:1.We proposed a new measure,namely feature disparity and algorithm of spatial dominant-feature co-location patterns to overcome the limitation that prevalence-based co-location mining method cannot discover the dominant relationship of features in co-location patterns.We also develop an efficient algorithm and corresponding pruning strategy to discover the spatial dominant-feature co-location patterns.The experiment results on the real datasets present that spatial dominant-feature co-location pattern is a kind of high-quality co-location pattern who considers the interaction of features in a co-location pattern.2.We propose a new measure,namely spatial occupancy to evaluate the representativeness of the spatial co-location instances in their neighborhoods,and discover the dominant co-location patterns.We also develop an efficient algorithm and corresponding pruning strategy to discover the dominant co-location pattern.The experiment results on real datasets present dominant co-location mining can effectively reduce the number of prevalent co-location patterns,and the dominant co-location pattern is a kind of high-quality pattern who considers the interaction of the co-location instances and the other instances in their neighborhoods.3.We propose a combined co-location pattern concept and corresponding interesting measures to overcome the limitation that single co-location pattern information cannot provide the global perspective of co-location pattern mining results.We also develop an efficient algorithm and a series redundancy limitation strategies to discover combined co-location patterns.The experiments results on real data sets show that the combined co-location pattern mining method can evaluate and analyze the patterns from a global perspective,and give a concise mode grouping result,which reveals that the combined co-location pattern is a kind of high-quality pattern who considers the interaction of the co-location patterns.