Uncertain Space With Rare Features Co-location Pattern Mining

Data mining is one of the most active branches of database researching, development and application. In recent years, along with the data collection technology's fast development, specially the GIS technology's widespread application, spatial data mining gradually becomes one of key studies. However, the data from many surveys may have the errors in them, the collected data is often not precise. Therefore to obtain the accurate results, it is very essential to do the research of uncertain data mining algorithm. The spatial co-location pattern mining is to find a group in which the features appear frequently together, and it is one of the most important domain of data mining. At present the uncertain spatial co-location pattern mining already have been made some outstanding progresses, however these achievements are not suited with all the situations. If there are rare features in the dataset, some frequent co-location patterns cannot be captured when we use the existed algorithms. Considering this condition, we advance a new approach of mining frequent spatial co-location patterns from uncertain dataset in which rare features exist.First, this paper introduces the concepts of uncertain data and uncertain data mining, and describes the researching situation together with significance at present.Second, the basic concepts of spatial co-location mining in cetain dataset and some classical alogthims such as Join-Based alogthim are introduced in this paper. Then this paper introduces the concept of rare feature and MaxPrune algorithm which is used to find spatial co-location patterns in uncertain dataset where there are rare features in it.Third, this paper introduces possible world, basic concepts of spatial co-location mining in uncertain dataset and the method of mining space co-location pattern from uncetain dataset in which the instances are associated with existential probabilities.Fourth, this paper expands MaxPrune algorithm to UMaxPrune algorithm which can be used to mine spatial co-location patterns in uncertain dataset containing rare features. In addtion, the shortcomings of the UMaxPrune algorithms are found out, and a new method called UWB algorithm which is based on weightedly expected paticipate rate are advanced. Using this method,we can successfully solve the shortcomings of UMaxPrune algorithm. According to the partial closure character of the weightedly expected participate rate, an improved algorithm which is named for UWP algorithm are advanced.Fifth, we make a lot of experiments through both the synthetic dataset and real dataset. It is proved that the method which is based on weightedly expected paticipate rate is correct and effective.Finally, the work of this paper are summarized, and future work of spatial co-location in the uncertain data set containing rare features is prospected.