Geographically weighted discriminant analysis of environmental conditions associated with Rift Valley fever outbreaks in South Africa

Abstract

Rift Valley fever (RVF) is a zoonotic arboviral infection that has occurred across Africa and parts of the Middle East. Geographically weighted discriminant analysis (GWDA) is a spatially-adaptive extension of traditional discriminant analysis (DA) which has rarely been applied to infectious disease epidemiology research. This study compares the classification performance of GWDA and traditional DA when used to distinguish between locations where livestock are at risk or are not at risk for acquiring RVF virus (RVFV) using 699 case reports of RVF (affecting 18,894 animals) from two outbreaks in South Africa in 2008–2009 and 2010–2011. GWDA produced better results than traditional DA for all bandwidth and kernel combinations. The best GWDA model correctly classified 96.6% of the original data versus 84.5% obtained with traditional DA. With GWDA, false positives decreased from 10.9% to 3.7%, and false negatives decreased from 19.9% to 3.2%.