Anomaly detection module for enterprise file integrations

In recent years, there has been an enormous increase in the variety of data storage and software development technologies. Due to the rapidly evolving and changing technologies, integration and diversity in collaborative organizations emerge as a fundamental problem. Enterprise file integration comes out as an effective solution in order to help integrate data between different business platforms. Enterprise file integration is mostly implemented through XML-based file transfers. The volume of file transfer usage may cause some anomalous symptoms during integrations. In this thesis, it is aimed to detect the anomalies in enterprise file integrations and a machine learning based anomaly detection solution is proposed for enterprise file integrations. For this purpose, the machine learning based anomaly detection models are trained by using Elliptic Envelope, Isolation Forest, One-Class Support Vector Machine and Local Outlier Factor models and their results are evaluated. The obtained results show that machine learning approaches are applicable for anomaly detection and could reach 89 percent of accuracy and 0.88 points of F1 score in performance measurements. As a secondary method for anomaly detection in enterprise file integrations, an ontology is developed to ensure semantic expressibility in the relevant domain. In the scope of evaluation, a comparison about the applicability of machine learning and Semantic Web in anomaly detection is presented.