Space Co-location Pattern Mining Uncertainty

With the rapid development in data gathering and processing, the uncertainty of thedata has been studied gradually in-depth. Uncertain data exists in many applications,such as economic, military, logistics, finance, telecommunications etc. It becomes moreand more important that how to effectively analyze the large amount of uncertain data inuncertain databases to discover the potential, valuable, and interesting information. Thespatial co-location pattern mining discovers the association relationships among thespatial events, which is an important branch in the field of the spatial data mining. Theresearch topic of discovering spatial co-location patterns from uncertain databases ischose to study in this thesis.First, the concepts,studying methods and the existing researches of the uncertainspatial data mining are introduced.Second, the concepts of co-location patterns and two classical algorithms whichare Join-Based algorithm and CPI-tree-Based algorithm are presented.Third, the problem of mining spatial co-location patterns from uncertain datawhose locations are described by probability density functions (PDF) is studied. It isshowed that the UJoin-Based algorithm, which generalizes the Join-Based algorithm tohandle uncertain instances, is very inefficient. The inefficiency comes from the factthat UJoin-Based computes expected distances (ED) between instances. For arbitraryPDF's, expected distances are computed by numerical integrations, which are costlyoperations. Various pruning methods are studied to avoid such expensive expecteddistance calculations. Experiments over synthetic data are conducted to evaluate theeffectiveness of our pruning techniques and illustrate the significance of our studies.Fourth, the problem of co-location patterns mining from uncertain data underpossible world model is studied. We consider spatial events whose instances areassociated with existential probabilities and gave a formal definition of spatialco-location patterns under possible world model. It is demonstrated that the algorithm for mining spatial co-location patterns under possible world model is time-consuming.An Optimization strategy is proposed to improve the efficiency of the algorithm.Through experiments on synthetic data, it is proved that the optimization strategy iseffective. Experiments on plant data show that our study over existential uncertain datais significant.At the last, the conclusion and future work are presented.