Outlier Detection Technology Research Based On Kernelized Function

In the modern computer Intelligence research area, Intelligence algorithms have been used in all walks of life. Outlier detection is an important part of data mining. It is usually been used for finding small model (comparing with clustering), that is to say, the data that specially different from other data. Outlier detection often apply in the field of telecommunications & credit card cheating, loan approval, weather forecast, financial industry and classification of customer, etc. Outlier detection develops from traditional statistic area to the widely research. Usually, the user firstly set up mathematical modeling by using statistic distribution, secondly using supposed modeling to decide whether the data is normal according to the data's distribution . Kernel function provide a system method for training learning machine. Using Kernel Function combines with Outlier detection can get better result.In this paper, we start from Kernel function and expend the original Intelligence algorithms. After referring some new ideas, based on the smallest sphere outlier detection and kernelized spatial depth outlier detection, the algorithms try to develop and expend. And then it is demonstrated by experiments. This paper is devoted to develop some ideas as follows :Firstly, fuzzy decision for outlier detection of kernelized spatial depth function. Using fuzzy decision method's decision veracity, the algorithm improve the outlier detection efficiency. The smallest sphere outlier detection is the method which can get satisfied efficiency. When these two algorithms merge the method of fuzzy decision, the outlier detection efficiency is improved.Secondly, the outlier detection of kernelized depth weighted, using distribution of weight can contribute to the improvement of detection efficiency. Besides, weight of the kernelized spatial depth function can get more exactly distance than normal weight. Finally, it can get more efficient outlier detection.Thirdly, it refers to one of the most important steps for outlier detection that is feature extraction.The kernelized spatial depth function merges with feature extraction. The Relief algorithm is the classical feature extraction method. It uses the kernelized spatial depth function algorithm in the feature extraction and improves the feature extraction result. And it is tested on the UCI dataset, the result have shown good status.Finally, In the paper, it used standard dataset to test these algorithm's efficiency. According to the experiment result, these algorithm based on kernel function has shown some improvement.