Towards Detection of Anomalies in Building Management Data

Abstract:

This paper aims at finding anomalies in multidimensional spatio-temporal data. We focus on building management data and describe a novel method for mining anomalies in those data. The main idea lies in building a model inductively from data and then in finding examples that are incorrectly classified by this model. Those exceptions are visualized.  We describe experiments with three tree learning algorithms for different classification tasks and discuss the results.