Using Machine Learning to Improve Public Health Surveillance in Low Income Countries

Abstract

Public health surveillance is a challenge across the globe. Unexpected disease outbreaks occur even though we have vast knowledge in the field of medicine. Also not being able to forecast the next set of diseases to be reported keeps our medical practitioners less prepared. This problem is heightened in low-income countries that cannot afford the infrastructure that comes with ubiquitous computing methods including Machine Learning on a national scale. There is also a notion that such computing approaches by huge companies such as Google and IBM are too expensive to implement. This research seeks to investigate if Machine Learning can be used in low-income countries to provide an effective public health surveillance regardless challenges in funding and availability of ubiquitous infrastructure. Disease Outbreak Detection and Disease Forecasting are the two branches of public health surveillance considered in this research. A prototype application called MATE is built based on Numenta’s implementation of Hierarchical Temporal Memory (HTM) Algorithms. Experiments were conducted with MATE proved 75% to 95% accuracy for forecasting next 20 diseases, and an instant recognition of anomalies for Disease outbreak detection.