Research And Application Of Outlier Detection Algorithm Based On Granular Computing

Outlier detection is an extremely important research in the field of data mining.Abnormal data behaves differently than most datas of dataset.Outlier detection is often used to remove anomalous data from dataset in preprocessing,the dataset of removing anomalous data often improves accuracy significantly in the statistics for supervised learning.Therefore,efficient outlier detection algorithms are crucial for data mining.At present,the research on outlier detection mainly includes methods based on statistics,clustering,classification and nearest neighbors.However,most of existing algorithms have shortcomings such as sensitive parameter selection,time-consuming and incomplete consideration of attributes.Granular computing is the multi-level and multi-perspective method.This topic integrates granular computing into outlier detection field,and multigranularity data outlier detection algorithms are proposed.The main contributions are as follows:1.In order to consider the neighbor distribution of the data,the concept of outlier index is introduced through the density of neighbor distribution,a multi-granularity outlier detection algorithm based on box plot and outlier index is proposed with the multi-granularity,the performance of the algorithm is verified about housing price prediction.2.Aiming at the problems of sensitive parameter selection and long timeconsuming of existing algorithms,a multi-granularity outlier detection algorithm based on Xm R control chart is proposed by combining the idea of control chart with granular computing.For each single granularity in the dataset,the algorithm uses the single-valuemoving range control chart(Xm R control chart)to construct X-map and m R-map from different perspectives,so as to mine outliers in the graph.3.The multi-granularity outlier detection algorithm based on Xm R control chart is applied to the field of software defect detection.The superiority of algorithm is verified by comparing with seven outlier detection algorithms(Isolation Forest,LOF,OCSVM,COF,CBLOF,HBOS and KNN).In this topic,two data outlier detection algorithms based on granular computing are proposed.The mean and standard deviation of different evaluation indicators are analyzed to verify the effectiveness of algorithms.Figure 34;Table 26;Reference 69...