Improving disaggregation models of malaria incidence by ensembling non-linear models of prevalence

• DOI: 10.1016/j.sste.2020.100357
• Publication Date: 2020-07-04T11:39:43Z
• Authors: Tim C.D. Lucas; Anita K. Nandi; Suzanne H. Keddie; Elisabeth G. Ches...; Rosalind E. Howes; Susan F. Rumisha; Rohan Arambepola; Amelia Bertozzi-V...; Andre Python; Tasmin L. Symons; Justin J. Millar; Punam Amratia; Penelope Hancock; Katherine E. Battle; Ewan Cameron; Peter W. Gething; Daniel J. Weiss
• Keywords: Predictive modelling; Ensemble Learning; Ensemble forecasting

Maps of disease burden are a core tool needed for the control and elimination malaria. Reliable routine surveillance data malaria incidence, typically aggregated to administrative units, is becoming more widely available. Disaggregation regression an important model framework estimating high resolution risk maps from data. However, aggregation incidence over large, heterogeneous areas means that these underpowered complex, non-linear models. In contrast, prevalence point-surveys directly linked local environmental conditions but not common in many world. Here, we train multiple non-linear, machine learning models on Plasmodium falciparum point-surveys. We then ensemble predictions with disaggregation uses incidences as response find using combine improves accuracy relative baseline model.