Tuesday, 5 August 2014

Malaria disease modeling

We are currently working on a number of ideas to improve the modeling results. These include
1) Finding a scientific and plausible way of adding human mobility information to disease models. Movement of an infected person from one area, can greatly contribute to a disease risk of another area more than proximity of the two locations. We are using call data records,CDR, for mining human mobility. One of the intermediate results in this work is shown in the figure below.

2) Processing of satellite imagery for both climatic proxies (like rainfall from Rainfall Estimates and normalized difference vegetation index - NDVI, temperature from land surface temperature - LST) and environmental factors (like topology from the digital elevation model - DEM. As part of this, we are also trying to visually understand the distribution of these factors to gain an insight into how well to model the Malaria disease over Uganda. Taking just a portion of Uganda in the figure below
To understand the effect of weather on vegetation over the selected area, we worked out the minimum and the maximum value of the Normalized Difference Vegetation Index, NDVI over the period of 2011.
3) Inferring unreported disease incidence counts from incomplete mTrac reports and environmental conditions. Malaria surveillance data in Uganda is always incomplete because of cases of non-reporting by some health facilities. Ensuring an adequate supply of antimalarial medication to rural health centers requires detailed, accurate and timely information about malaria cases. Thus we want to use the spatio-temporal correlations of both reporting and environmental conditions to infer unreported counts of malaria. mTrac, a project supported by UNICEF, is a simple SMS data collection service, which allows health centers to text their weekly reports.

This work is a joint collaboration between the Machine Learning Lab at the University of Sheffield and AI Lab at Makerere University.