Outlier detection by using an artificial immune system-based algorithm

Outlier detection, which is applied successfully in many fields such as credit card fraud detection, network intrusion detection, extracting errors from multi-dimensional data and easing the burden of big data systems, is a data mining methodology which follows the principle of textually rich but information poor approach.It is very difficult to realize the outlier detection in multi-dimensional data sets. Although there are many algorithms that may be effective in outlier detection, many of these algorithms are impractical for big and large-sized data sets. As the number of features and samples in the data increases, the success of the data mining methods in outlier detection decreases. In this dissertation, the artificial immune system-based algorithm that can realize outlier detection in data sets easily and produce solutions that are more effective than other methods was used. The algorithm was applied to the three data sets two of which are taken from two real-life data sets and one of which is made up of artificial data set. The performance of the algorithm was compared with the performances of the k-Nearest Neighbour Algorithm, Distance- Based Outlier Detection Algorithm and Box Plot method in outlier detection. When the results were compared, it was found out that Artificial Immune System-Based Algorithm gives better results and works with a lower error rate than the other methods used in outlier detection.