Spatiotemporal patterns and machine-learning prediction of malaria risk across Africa, 2000–2024

Malaria is currently a challenge in Africa, where transmission is shaped by interacting climatic, demographic and socioeconomic conditions. Improved prediction could support earlier identification of high-risk settings and more targeted use of control resources. This study evaluated multiple machine-learning approaches for malaria-risk prediction across Africa and examined the variables driving model predictions. Malaria burden data for 2000–2024 from the Malaria Atlas Project (MAP) were combined with climate indicators from NASA POWER and demographic, socioeconomic and health indicators from the World Bank. Descriptive analyses covered 47 African countries, while the harmonised machine-learning dataset comprised 1,056 country-year observations from 44 countries. Six classification algorithms were used including, logistic regression, decision tree, Random Forest, support vector machine (SVM), gradient boosting machine (GBM) and XGBoost. Model development used leakage-free expanding-window temporal cross-validation, while 2021–2024 was retained as an independent out-of-time test period. Uncertainty was assessed with country-cluster bootstrap confidence intervals model calibration and internal grouped-country performance were also examined. Performance was assessed using accuracy, balanced accuracy, sensitivity, specificity, F1-score, Receiver Operator Curves (ROC)-Area Under Cureve (AUC), PR-AUC and Brier score. Across 2000–2024, the dataset contained 5.63 billion reported clinical malaria cases and 17.78 million malaria deaths. Cases increased from 204.3 million in 2000 to 244.7 million in 2024, while deaths declined from their early-period peak. Random Forest had the largest mean temporal cross-validated ROC-AUC (0.968). In the independent test period, the four principal machine-learning models showed similar discrimination: ROC-AUC was 0.955 for SVM, 0.950 for Random Forest, 0.949 for GBM and 0.943 for XGBoost, and pairwise confidence intervals for their ROC-AUC differences included zero. Random Forest achieved 0.903 accuracy and 0.902 balanced accuracy. Performance was lower when entire countries were held out, with pooled ROC-AUCs of 0.874 for Random Forest and 0.870 for XGBoost. Model interpretation showed that climatic and socioeconomic variables contributed jointly to predictions, with wind speed, lagged temperature, life expectancy, age dependency and unemployment among the leading predictors. Machine-learning models may provide useful malaria risk prediction from routinely available climate and socioeconomic indicators. However, geographic transferability is less certain than temporal prediction within represented countries. Further work should use finer spatial and temporal data, external validation and locally relevant intervention indicators before operational use.

Authors

Institutions

Publication Details

Journal
Discover Artificial Intelligence
Published
2026-11-13
DOI
https://doi.org/10.1007/s44163-026-02748-1
Primary Topic
Malaria Research and Control
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Spatiotemporal patterns and machine-learning prediction of malaria risk across Africa, 2000–2024

Sylvester Dassah, Samuel Ayetibo Ofori, Godfred Agongo, James Abugri et al.
2 links
Discover Artificial Intelligence
Malaria Research and Control
article

Spatiotemporal patterns and machine-learning prediction of malaria risk across Africa, 2000–2024

Sylvester Dassah, Samuel Ayetibo Ofori, Godfred Agongo, James Abugri, Agyapong Kofi Addo, Joseph Okaya
article en

Abstract

Malaria is currently a challenge in Africa, where transmission is shaped by interacting climatic, demographic and socioeconomic conditions. Improved prediction could support earlier identification of high-risk settings and more targeted use of control resources. This study evaluated multiple machine-learning approaches for malaria-risk prediction across Africa and examined the variables driving model predictions. Malaria burden data for 2000–2024 from the Malaria Atlas Project (MAP) were combined with climate indicators from NASA POWER and demographic, socioeconomic and health indicators from the World Bank. Descriptive analyses covered 47 African countries, while the harmonised machine-learning dataset comprised 1,056 country-year observations from 44 countries. Six classification algorithms were used including, logistic regression, decision tree, Random Forest, support vector machine (SVM), gradient boosting machine (GBM) and XGBoost. Model development used leakage-free expanding-window temporal cross-validation, while 2021–2024 was retained as an independent out-of-time test period. Uncertainty was assessed with country-cluster bootstrap confidence intervals model calibration and internal grouped-country performance were also examined. Performance was assessed using accuracy, balanced accuracy, sensitivity, specificity, F1-score, Receiver Operator Curves (ROC)-Area Under Cureve (AUC), PR-AUC and Brier score. Across 2000–2024, the dataset contained 5.63 billion reported clinical malaria cases and 17.78 million malaria deaths. Cases increased from 204.3 million in 2000 to 244.7 million in 2024, while deaths declined from their early-period peak. Random Forest had the largest mean temporal cross-validated ROC-AUC (0.968). In the independent test period, the four principal machine-learning models showed similar discrimination: ROC-AUC was 0.955 for SVM, 0.950 for Random Forest, 0.949 for GBM and 0.943 for XGBoost, and pairwise confidence intervals for their ROC-AUC differences included zero. Random Forest achieved 0.903 accuracy and 0.902 balanced accuracy. Performance was lower when entire countries were held out, with pooled ROC-AUCs of 0.874 for Random Forest and 0.870 for XGBoost. Model interpretation showed that climatic and socioeconomic variables contributed jointly to predictions, with wind speed, lagged temperature, life expectancy, age dependency and unemployment among the leading predictors. Machine-learning models may provide useful malaria risk prediction from routinely available climate and socioeconomic indicators. However, geographic transferability is less certain than temporal prediction within represented countries. Further work should use finer spatial and temporal data, external validation and locally relevant intervention indicators before operational use.

Discover Artificial IntelligenceVol. 6(1)
Navrongo Health Research Centre (GH), C.K. Tedam University of Technology and Applied Sciences (GH), Akenten Appiah-Menka University of Skills Training and Entrepreneurial Development (GH), University for Development Studies (GH)
Good health and well-being
Malaria Research and Control
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon