Modeling the spatiotemporal dynamics of malaria disease outbreaks in southwestern Ethiopia using deep learning

Malaria remains a major public health challenge in Sub-Saharan Africa, with climate change intensifying its transmission. In Ethiopia, 60% of the population faces the risk of malaria, which is highly seasonal, unstable, and closely linked to environmental factors. In particular, the South-west region experiences a unique combination of complex socio-economic challenges and vulnerability to malaria outbreaks. This study aims to analyze the spatiotemporal dynamics of malaria disease outbreaks in southwest Ethiopia and develop a predictive model to assist the management and prevention of malaria outbreaks. This study used retrospective data from the Ethiopian Public Health Institute, the Public Health Emergency Management, the Ministry of Health, the Jimma Meteorological Office, and the Jimma Zone Health Office, incorporating information on malaria case counts, meteorological factors, elevation, and health infrastructure. The dataset spans 38 woredas (districts) across five zones in Southwest Ethiopia from 2014 to 2020. The analysis involved three key approaches: (i) spatial analysis to identify geographic hotspots using historical malaria data alongside geospatial information, (ii) time series analysis to examine the malaria incidence, trends, and seasonality, and (iii) predictive modeling using a Long Short-Term Memory (LSTM) deep learning model and the Seasonal Autoregressive Integrated Moving Average with Exogenous Variables (SARIMAX) statistical model to forecast future outbreaks. Our results reveal that, from 2014 to 2020, a general reduction in malaria cases was observed, alongside distinct spatiotemporal patterns in malaria incidence across Southwest Ethiopia. Over time, a shift in malaria distribution was evident, with some areas experiencing reductions, while others, particularly the Bench Maji Zone, saw increases after 2017. The LSTM model demonstrated high accuracy in forecasting malaria outbreaks, achieving an R-squared value of 0.98 and a mean squared error (MSE) of 73.8, effectively capturing trends and fluctuations. In comparison, the SARIMAX model showed moderate accuracy, with an R-squared of 0.68 and a higher MSE of 1517, capturing general trends but proving less precise in predicting outbreaks. This study highlights the significant influence of environmental factors, such as proximity to rivers and favorable mosquito breeding conditions, on the persistence of malaria hotspots in the eastern and western parts of the Jimma, Dawro, and Bench Maji Zones. Incorporating predictive models into malaria control strategies can improve resource allocation, strengthen public health interventions, and support adaptive responses to environmental and socio-economic changes. Overall, the findings highlight the vital roles of environmental factors and healthcare infrastructure in influencing malaria patterns, stressing the need for timely, targeted interventions in high-burden regions.