Anomalies in link mining based on mutual information

The literature review found surprisingly low utilisation of mutual information in detecting anomalies in various domains, however no such study in link mining was found. This research is intended to fill the gap in link mining domain, although it has been widely used in other areas of data analysis. The current study is a first-step exploration of a new method that uses mutual information based measures to interpret anomalies and link strength between individual anomalies in a given dataset. Anomalies detection, which is the focus of this research, is concerned with the problem of finding non-conforming patterns in datasets. This study describes a novel approach to measure the amount of information shared between any random anomaly variables. Two types of data were used to evaluate the proposed approach: proof of concept data in case study 1 and citation data in case study 2. The CRISP data mining methodology was updated to be applicable for link mining study. The proposed mutual information approach to provide a semantic investigation of the anomalies and the updated methodology can be used in other link mining studies such as fraud detection, network intrusion detection and law enforcement areas which are expected to grow.