Measuring seasonal population dynamics
Understanding seasonal population dynamics, including how people move through landscapes at different times of year, is fundamental to understanding malaria transmission. Agricultural cycles, religious gatherings, school terms, and economic migration all shape when and where people are exposed to malaria risk.
This research stream integrates multiple data sources including smartphone location histories, Meta Activity Space Map data, WorldPop population models, and point-of-interest data to characterise seasonal shifts in population distribution and mobility. These dynamics are analysed within a malaria transmission modelling framework to improve the precision and timeliness of risk mapping.
Fine-scale transmission mapping
In Zanzibar, cases are attributed to place of residence, but a majority of transmission occurs outdoors, away from home. Construction sites, agricultural fields, and evening gathering spots are likely true transmission foci, yet they go unmapped. This research identifies and characterises true micro-transmission sources and sinks by combining individual-level smartphone location data from malaria-positive cases with parasite genomics, point-of-interest data, and remote sensing.
By tracking where cases spent time during peak vector biting hours (18:00–06:00), and applying spatial clustering algorithms to identify hotpops and hotspots, we produce fine-scale maps that enable precisely targeted interventions. The AI centred, autoML driven approach to identifying transmission sources and hotspots is more resource-efficient than blanket indoor approaches.
Importation pathway analysis
Over 60% of malaria cases detected in Zanzibar are linked to importation from mainland Tanzania, yet importation is not routinely incorporated into intervention planning. This research characterises and predicts regional-scale importation pathways, using aggregated mobile phone data, call records, and parasite genomics to identify the mainland source regions, travel corridors, entry points, and destination sinks that drive sustained transmission.
Seasonal patterns matter too: travel peaks in May and October coincide with Zanzibar's rainy seasons, while the Christmas period brings the highest volumes of visitors. Understanding these dynamics enables proactive screening, targeted messaging, and early warning systems calibrated to real importation risk.
AutoML integration for near real-time analysis
Malaria control programmes now have access to more data than ever, but the scale and complexity of integrating diverse streams often exceeds local analytical capacity. The MalarAI AutoML pipeline continuously processes and harmonises case surveillance, human mobility, parasite genomics, climate, environmental, and point-of-interest data to identify transmission risks, classify micro-sources and sinks, and generate actionable intervention guidance in near real-time.
The pipeline uses an ensemble of modelling approaches including gradient-boosted trees, random forests, neural networks, and spatiotemporal Bayesian models with automated hyperparameter tuning and continuous updating. Outputs include a web-based risk dashboard, short-term case forecasts, importation risk estimates by entry point, and automated daily briefings for ZAMEP staff. The system is designed for local ownership: modular, maintainable, and not dependent on external expertise.