Characterising selected algorithms used for outlier detection and developing improved combinations

Abstract

Data analysis is becoming very important, and outlier detection is one major newlinetype of data analysis problem. Outliers are points that are different from the remaining newlinedataset. Outliers contain useful information regarding abnormal characteristics of the newlinesystems and entities that impact the data generation process. Recognition of such newlineabnormal characteristics can lead to useful application-specific insights, especially in newlineintrusion detection and fraud detection cases. newlineAlthough there are many outlier identification approaches, they do not all newlineperform equally well on all types of datasets. Therefore, it can be assumed that certain newlinetools and techniques of outlier detection will perform well with datasets having newlinecertain characteristics only. The main objective of this work is to explore and find a newlinematch between the dataset characteristics and tools or techniques of outlier detection newlinethat perform well with a given dataset type. Algorithm developers are interested in newlinecreating new outlier detection algorithms. Algorithm users seek to identify and use the newlinemost suitable outlier detection algorithm for their problem dataset. However, newlinematching outlier detection algorithms and the type of dataset has to be done before newlineselecting a specific outlier detection algorithm. Information regarding match between newlineAlgorithm and Data set type is not available. This research is devoted to developing a newlinemethodology and finding match between data set type and the most suitable outlier newlinedetection algorithm. This approach assumes that users are aware of their dataset type. newline