Geospatial and machine learning approaches for malaria risk mapping in flood-prone districts: Implications for public health decision-making.

OBJECTIVES

Malaria remains a major public health concern in flood-prone districts where environmental vulnerability and weak health infrastructure exacerbate transmission risks. This study develops an integrated geospatial framework for malaria risk zonation by combining multi-criteria decision-making and machine learning.

STUDY

DESIGN

A geospatial modeling study integrating environmental, meteorological, and socio-demographic datasets with decision analysis and machine learning.

METHODS

Eleven predictor variables, including elevation, land surface temperature (LST), rainfall, slope, drainage density, humidity, flood inundation, population, proximity to roads and health facilities, and land use/land cover (LULC), were processed to generate hazard, vulnerability, and elements-at-risk (EAR) layers. Two approaches were employed: (i) the Analytical Hierarchy Process (AHP) integrating hazard, vulnerability, and EAR through weighted overlays; and (ii) a Random Forest (RF) model trained with 250 union council-level malaria Test Positivity Rate (TPR) records from 2014 to 2024.

RESULTS

The RF model achieved 93.3% accuracy, 0.95 precision, 0.93 recall, and a Kappa coefficient of 0.86, confirming strong predictive performance. AHP identified 69.2% of the area as moderate risk and 13.8% as high risk, while RF produced localized hotspots with higher spatial resolution. Flood proximity, LST, rainfall, and LULC were dominant predictors. Validation against 2022 malaria outbreak data showed strong spatial agreement.

CONCLUSIONS

The dual-model framework, integrating hazard, vulnerability, and EAR layers with data-driven validation, demonstrates practical utility for climate-resilient malaria control. The methodology is transferable to other disaster-prone regions for targeted interventions and resource allocation.