Unsupervised Consensus-Based Anomaly Detection for Spatiotemporal Malaria Incidence in Ghana

This study introduces a consensus-based, unsupervised machine learning framework to identify atypical malaria transmission patterns in Ghana using monthly surveillance data from 2014 to 2023. By integrating multiple anomaly detection algorithms, the researchers aimed to reduce model-specific bias and capture complex, non-linear transmission dynamics that traditional threshold-based surveillance systems often miss. The framework processes regional data to flag month-region combinations that deviate significantly from historical norms, such as unexpected surges or persistent off-season transmission.

The analysis demonstrates that malaria anomalies are not randomly distributed but are highly structured across space and time. The Ashanti and Northern regions were identified as hotspots for recurrent anomalies. A critical insight is the distinction between 'anomaly burden' and 'anomaly frequency': while Tamale recorded the highest total case burden during anomalous periods, a cluster of districts in the Ashanti region exhibited the highest frequency of persistent anomalous behavior. Furthermore, anomalous months were found to be statistically distinct from normal periods, showing significantly higher case counts and larger deviations from seasonal baselines.

Traditional malaria surveillance often focuses on aggregate case counts, which can obscure short-lived or localized transmission events. By distinguishing between high-prevalence areas and areas with unusual transmission behavior, this framework provides public health authorities with a more nuanced tool for resource allocation. It enables the prioritization of epidemiological investigations in areas where transmission dynamics are shifting, potentially allowing for earlier intervention before these anomalies escalate into broader public health crises.