Efficient and Accurate Non-exhaustive Pattern-Based Change Detection in Dynamic Networks
Pattern-based change detectors (PBCDs) are non-parametric unsupervised change detection methods that are based on observed changes in sets of frequent patterns over time. In this paper we study PBCDs for dynamic networks; that is, graphs that change over time, represented as a stream of snapshots. Accurate PBCDs rely on exhaustively mining sets of patterns on which a change detection step is performed. Exhaustive mining, however, has worst case exponential time complexity, rendering this class of algorithms inefficient in practice. Therefore, in this paper we propose non-exhaustive PBCDs for dynamic networks. The algorithm we propose prunes the search space following a beam-search approach. The results obtained on real-world and synthetic dynamic networks, show that this approach is surprisingly effective in both increasing the efficiency of the mining step as in achieving higher detection accuracy, compared with state-of-the-art approaches.
KeywordsChange detection Pattern mining
- 2.Bailey, J.: Statistical measures for contrast patterns. In: Contrast Data Mining: Concepts, Algorithms, and Applications, pp. 13–20. CRC Press (2013)Google Scholar
- 6.Koh, Y.S.: CD-TDS: change detection in transactional data streams for frequent pattern mining. In: 2016 International Joint Conference on Neural Networks, IJCNN 2016, 24–29 July 2016, Vancouver, BC, Canada, pp. 1554–1561 (2016)Google Scholar
- 8.van Leeuwen, M., Siebes, A.: Streamkrimp: detecting change in data streams. In: Proceedings of European Conference on Machine Learning and Knowledge Discovery in Databases (Part I), ECML/PKDD 2008, 15–19 September 2008, Antwerp, Belgium, pp. 672–687 (2008)Google Scholar
- 10.Padillo, F., Luna, J.M., Ventura, S.: Subgroup discovery on big data: pruning the search space on exhaustive search algorithms. In: 2016 IEEE International Conference on Big Data, BigData 2016, 5–8 December 2016, Washington DC, USA, pp. 1814–1823 (2016)Google Scholar