Detection and explanation of exceptional values in a datamining/OLAP framework



In this presentation, we propose an extension of the datamining framework with causal diagnosis, offering the possibility to automatically generate explanations and diagnostics to support business decision tasks. This functionality can be provided by extending the conventional datamining system with an explanation formalism, which mimics the work of business decision makers in diagnostic processes. The central goal is the identification of specific knowledge structures and reasoning methods required to construct computerized explanations from data and business models. Therefore, we present a general methodology for automated diagnosis. Specifically, a look-ahead algorithm is proposed that deals with so-called cancelling-out effects, which are a common phenomenon in financial data sets. Firstly, the extended methodology is tested on a case-study conducted for Statistics Netherlands involving the comparison of financial figures of firms in the Dutch retail branch. Secondly, we tested the model on a more complex data set with multidimensional (OLAP) sales data for the automated detection and explanation of exceptional values. The analyses are performed with a diagnostic software application which implements our theory of explanation. More information: contact