Improving the Fusion of Outbreak Detection Methods with Supervised Learning
- 121 Downloads
Epidemiologists use a variety of statistical algorithms for the early detection of outbreaks. The practical usefulness of such methods highly depends on the trade-off between the detection rate of outbreaks and the chances of raising a false alarm. Recent research has shown that the use of machine learning for the fusion of multiple statistical algorithms improves outbreak detection. Instead of relying only on the binary outputs (alarm or no alarm) of the statistical algorithms, we propose to make use of their p-values for training a fusion classifier. In addition, we also show that adding contextual features and adapting the labeling of an epidemic period may further improve performance. For comparison and evaluation, a new measure is introduced which captures the performance of an outbreak detection method with respect to a low rate of false alarms more precisely than previous works. We have performed experiments on synthetic data to evaluate our proposed approach and the adaptations in a controlled setting and used the reported cases for the disease Salmonella and Campylobacter from 2001 until 2018 all over Germany to evaluate on real data. The experimental results show a substantial improvement on the synthetic data when p-values are used for learning. The results on real data are less clear. Inconsistencies in the data appearing under real conditions make it more challenging for the learning approach to identify valuable patterns for outbreak detection.
KeywordsOutbreak detection Fusion methods Stacking Syndromic surveillance
This work was supported by the Innovation Committee of the Federal Joint Committee (G-BA) [ESEG project, grant number 01VSF17034]. We thank our project partners the Health Protection Authority of Frankfurt, the Hesse State Health Office and Centre for Health Protection, the Hesse Ministry of Social Affairs and Integration, the Robert Koch-Institut, the Epias GmbH and the Sana Klinikum Offenbach GmbH who provided insight and expertise that greatly assisted the research. Especially, we thank Linus Grabenhenrich, Alexander Ulrich, Theresa Kocher, Madlen Schranz, Sonia Boender and Birte Wagner from the Robert Koch-Institut for their valuable feedback, that substantially improved the manuscript, and for providing us the data for the evaluation.
- 1.Jafarpour, N., Precup, D., Izadi, M., Buckeridge, D.: Using hierarchical mixture of experts model for fusion of outbreak detection methods. In: Annual Symposium Proceedings 2013, pp. 663–669, November 2013Google Scholar
- 4.Kulessa, M., Loza Mencía, E., Fürnkranz, J.: Improving outbreak detection with stacking of statistical surveillance methods. In: Workshop Proceedings of epiDAMIK: Epidemiology meets Data Mining and Knowledge Discovery (held in conjunction with ACM SIGKDD 2019) (2019). Also as preprint arXiv:1907.07464