Feasibility study of applying descriptive ILP to large geographic databases

This paper discusses a case study in which the aim is to discover regularities and anomalies in large databases containing geographic data, to improve and maintain the overall data quality. The application of Inductive Logic Programming (ILP) and descriptive ILP in particular to this case is discussed and motivated. In an experiment on real-world data, a classical descriptive ILP algorithm (WARMR) is applied to the hamlet of Beggen, Luxemburg, to mine for rules describing regularities. The algorithm adopts the setting of learning from interpretations. In a next stage, the violating interpretations of the rules could be traced to identify candidate anomalies. A rule export module was set up to feed the results to a rule checking engine for further validation of this experiment. Finally, the results are discussed, the feasibilities of the system used in the case study are assessed and possibilities w.r.t. a larger scale application of the experiment are discussed.