A Case Study of Cluster-based and Histogram-based Multivariate Anomaly Detection Approach in General Ledgers

The rapid development of financial markets results in data variability and unpredictability. Anomaly detection in financial data is a very important issue. Finding anomalies can result in error reduction and corrections in due time. The main aim of this research was to find anomalies in general ledgers of a real company in Bosnia and Herzegovina. Anomalies are defined as input errors of accountants. Main concepts of anomaly detection are defined, a summary of the current progress is given, and challenges of future work are presented. Cluster-based and histogram-based anomaly detections were performed on a real-life dataset of a microcredit organization. Results of algorithms were presented, as well as results achieved using synthetic data.