Seminář: Anomaly Detection on Numerical Financial Data

The number of open government budget and spending datasets published on the Web of Data is constantly increasing. Hence, being able to automatically analyze such datasets is essential. When analyzing financial data we are interested in unusual values, i.e. anomalies, like a city receiving 10 times more EU subsidies than other cities of comparable size. These can be detected using Outlier Detection techniques. However, if the city has a high unemployment rate the money might be justified. So, in our approach, we first collect further information features) by following links in the data set to LOD sources. We then systematically apply several outlier detection steps on a subpopulation lattice grouping comparable amounts of spent money together. Finally, the resulting outlier scores are combined and presented to the user together with possible explanations extracted from the respective subpopulation constraints. The approach will be tested on real-world open budget and spending data sets.