If you need to read two review articles about outlier detection, the first one is…
Outlier Detection: A Survey
The first one, Outlier Detection: A Survey, is written by Chandola, Banerjee and Kumar. They define outlier detection as the problem of “[…] finding patterns in data that do not conform to expected normal behavior“. After an introduction to what outliers are, authors present current challenges in this field. In my experience, non-availability of labeled data is a major one.
If you need to read two review articles about outlier detection, the first one is…
Outlier Detection: A Survey
The first one, Outlier Detection: A Survey, is written by Chandola, Banerjee and Kumar. They define outlier detection as the problem of “[…] finding patterns in data that do not conform to expected normal behavior“. After an introduction to what outliers are, authors present current challenges in this field. In my experience, non-availability of labeled data is a major one.
The authors proposes three types of supervisions. In supervised outlier detection we make the assumption that labeled data are available. Semi-supervised outlier detection assumes that only one class of labeled data is available. Techniques which models normal instances as the only class are more popular (since normal instances are easier to obtain). The third approach, unsupervised outlier detection, is the most widely used one. The paper continues by describing three types of outliers. Authors then describes several applications of outliers detection in areas such as intrusion detection, fraud detection, industrial damage detection, image processing, etc.
Techniques used for outlier detection are then described. It is surprising to read that most data mining techniques can be applied to the task of outlier detection. For example: neural networks, SVM, rule-based, clustering, nearest neighbors, regression, etc. The articles continues with several other techniques. Authors also describe ways to evaluate results of outlier detection with false positive, false negative and ROC curve. To be noted the 19 pages (!) of references to other articles in the field. One of their main conclusions is that “[…] outlier detection is not a well-formulated problem“. It is your job, as a data miner, to formulate it correctly.
Link to Outlier Detection: A Survey
