This project is being carried out in collaboration with the Faculty of Tropical Medicine and the Center of Excellence in Biomedical and Public Health Informatics (BIOPHICS) at Mahidol University. It involves development of spatio-temporal Bayesian networks to effectively integrate crowd-sensed and authority-sensed data concerning infectious disease. The techniques promise to help provide rapid and reliable outbreak detection, prediction, and evaluation of control strategies. Initial focus of our work has been on malaria which continues to pose a serious public health threat in tropical countries such as Thailand and its neighbors, particularly in remote areas. Work on dengue is planned in the near future.

Accurate use of crowdsourced data has typically relied on data volume, applying simple statistics to the data and assuming that unreliable reports constitute noise at a manageable level. Little work has addressed effectively using crowdsourced data in areas where data volume is considerably lower, such as remote areas in developing countries [1]. We are seeking to demonstrate how spatio-temporal Bayesian network models may be used to represent physical, process, and other constraints for reliable data interpretation in such cases.

Most traditional methods for disease surveillance use fairly coarse aggregate statistics, comparing numbers of cases against a null hypothesis. Spatial heterogeneity is rarely accounted for [2]. Recent work on ecological niche modeling has sought to achieve fine spatial resolution and to account for heterogeneity and complexity. While the niche modeling approach can provide rich models for data fusion and interpretation, creating detailed niche models manually for each situation assessment task is far too laborious to be practical. While suggestions have been made that we need to have “libraries” of models to draw upon [3], such an approach to niche modeling has yet to be realized. Fortunately, general methods exist for creating context-specific Bayesian network models from libraries of generic model fragments [4, 5, 8]. Fragments can be represented in the form of probability logic, which permits expression of general schemas through the use of quantified variables. The variables are then instantiated and the fragments assembled in the context of a particular problem, much like sentences in a first-order logic knowledge base are composed by a theorem prover. Some recent work has also developed techniques for integrating Bayesian networks with Geographic Information Systems [6, 7] and commercial software is emerging.

The techniques we are developing will support prediction and evaluation of control strategies by integrating information from case reports, crowdsourced syndromic reports, as well as environmental, socio-economic, and demographic information. The approach consists of learning Bayesian network model fragments from data and storing them in a model knowledge base. Using the information stored in a GIS, the fragments are assembled into tailored, location specific spatio-temporal Bayesian networks. Data for the models comes from the Thai national electronic malaria information system (eMIS) built and maintained by BIOPHICS. Modeling is at the village level, initially focusing districts in Thailand along the border with Myanmar.

Specific research issues we are examining include development of:

- high spatial and temporal resolution Bayesian network models for malaria prediction [9];
- ensemble techniques for malaria prediction [10];
- a probabilistic data integration framework to link predictive and diagnostic models in order to increase the accuracy of interpretation of crowdsourced data and facilitate its integration into predictive models;
- techniques for spatial clustering at multiple levels of aggregation in order to be able to make principled informed tradeoffs between spatial resolution and prediction model accuracy;
- techniques to generate Bayesian network models from libraries of fragments, particularly to model spatial disease autocorrelation; and
- techniques to extract physical features from maps and make use of them in automatically configuring prediction models.

**References**

[1] Haddawy, P., Frommberger, L., Kauppinen, T., De Felice, G., Charkratpahu, P., Saengpao, S., Kanchanakitsakul, P., Situation awareness in crowdsensing for disease surveillance in crisis situations, Proceedings of the Seventh International Conference on Information and Communications Technologies and Development (ICTD'15), Singapore, 2015.

[2] Robertson, C., Nelson, T.A., MacNab, Y.C., Lawson, A.B. Review of methods for space-time disease surveillance, Spatial and Spatio-temporal Epidemiology, 1: 105-116, 2010.

[3] Peterson, A.T., Martinez-Campos, C., Nakazawa, Y., Martinez-Meyer, E. Time-specific ecological niche modeling predicts spatial dynamics of vector insects and human dengue cases, Transactions of the Royal Society of Tropical Medicine and Hygiene, 99: 647 – 655, 2005.

[4] Ngo, L. and Haddawy, P. Answering Queries from Context-Sensitive Probabilistic Knowledge Bases, Theoretical Computer Science, 171(1-2):147- 177, 1997.

[5] Laskey, K.B. MEBN: A Language for First-Order Bayesian Knowledge Bases, Artificial Intelligence, 172(2-3): 140-178. 2008.

[6] Laskey, K.B., Wright, E.J., da Costa P.C.G. Envisioning uncertainty in geospatial information, International Journal of Approximate Reasoning, 51:209-223, 2010.

[7] Johnson, S., Low-Choy, S., Mengersen, K. Integrating Bayesian networks and Geographic Information Systems: Good practice examples. Integrated Environmental Assessment and Management, 8(3): 473 – 479, 2012.

[8] Ngo, L., Haddawy, P., Helwig, J., Krieger, B. Efficient Temporal Probabilistic Reasoning Via Context-Sensitive Model Construction. Computers in Biology and Medicine, 27(5):453-476, 1997.

[9] P. Haddawy, R. Kasantikul, A.H.M. Imrul Hasan, C. Rattanabumrung, P. Rungrun, N. Suksopee, S. Tantiwaranpant, N. Niruntasuk, Spatiotemporal Bayesian Networks for Malaria Prediction: Case Study of Northern Thailand, In Studies in Health Technology and Informatics, vol 228: Exploring Complexity in Health: An Interdisciplinary Systems Approach, pp 773-777, Aug 2016.

[10] A.H.M. I. Hasan and P. Haddawy, Integrating ARIMA and Spatiotemporal Bayesian Networks for High Resolution Malaria Prediction, In Frontiers in Artificial Intelligence and Applications, vol 285: Proc. European Conference on Artificial Intelligence (ECAI 2016), Aug 2016.

Theme by Danetsoft and Danang Probo Sayekti inspired by Maksimer