Abstract:
An anomaly scoring algorithm assigns a score to an instance that provides a large value for an outlier in order to help detecting anomalies within a dataset. In 2013, one of the parameter-free techniques called the order difference distance outlier factor was proposed (OOF). OOF was computed using the ordered difference distance derived from the distance matrix sorted in each row, before calculating the difference. The minimum distance was included to avoid false detection but it also decreased the score of anomalies forming a small cluster. To avoid the use of the minimum distance, the new technique is proposed base on the ordered difference distance considering along the angle which is called the acute angle order difference distance outlier factor (AOF). The various collections of synthesized datasets are experimented to exhibit the performance of AOF. Moreover, to improve the detection rate of AOF, the enhanced version of AOF is also propose in this thesis.