Y1 Progress report | Jacob Ageykum

Harnessing Community Resilience and Machine Learning for Adaptive Malaria Control amid Climate Change in the Upper West Region of Ghana

The first year of the research project focused on establishing the scientific, ethical, technical, and stakeholder foundations necessary to develop an artificial intelligence-driven, climate-informed malaria early-warning framework for Ghana. Over the reporting period, substantial progress was achieved across ethical approvals, stakeholder engagement, data acquisition and processing, community engagement, statistical analysis, machine learning model development, scientific collaborations, and research dissemination. These activities have strengthened the project's interdisciplinary nature by integrating climate science, epidemiology, artificial intelligence, and community knowledge to improve malaria surveillance and public health decision-making.

Ethical Clearance and Research Preparation

A major milestone during the first year was the successful completion of the ethical clearance process. Preparations for ethical approval began with consultations to understand the application requirements and to develop research instruments, including household survey questionnaires and focus group discussion guides. These instruments underwent several rounds of review with my supervisor to ensure scientific rigour, methodological appropriateness, and compliance with ethical standards. Following submission via the institutional online ethics platform, ethical approval was granted, allowing the project to proceed with field-based community engagement and data collection.

Stakeholder Engagement and Partnerships

Stakeholder engagement formed another critical component of the project's first year. Early consultations were held with the Ghana Health Service (GHS), the Ghana Meteorological Agency (GMet), and regional health authorities to introduce the project objectives, establish collaborations, and secure access to the required datasets. Meetings and discussions with these institutions facilitated the acquisition of malaria surveillance data and meteorological observations, and secured agreements to supplement station observations with satellite and reanalysis climate products such as CHIRPS and ERA5. These collaborations established a strong institutional foundation that will support future implementation and sustainability of the research outputs.

Data Collection, Processing, and Preliminary Analysis

Considerable effort was devoted to assembling and preparing the datasets required for predictive modelling. Malaria case records obtained from the Ghana Health Service were cleaned, quality-controlled, and converted into formats suitable for machine-learning applications. Climate datasets from GMet and complementary remotely sensed products were processed and harmonised similarly. Given the project's multidimensional nature, substantial preprocessing was required to integrate climate, environmental, and health datasets with varying temporal and spatial resolutions. This process included quality assurance, temporal alignment, spatial matching, handling missing observations, and feature engineering to ensure that the resulting database could support advanced predictive modelling.

Alongside data preparation, preliminary statistical analyses were undertaken to investigate the relationships between climate variability and malaria incidence. Exploratory analyses examined the influence of rainfall, temperature, humidity, and other environmental variables on malaria transmission patterns across the study region. Time-series analyses, trend assessments, correlation analyses, and investigations of lagged climate effects were conducted to identify key environmental drivers of malaria transmission. These analyses provided valuable insights into seasonal disease dynamics and informed the selection of predictor variables for subsequent machine learning model development.

Machine Learning Model Development

An important achievement during the first year was the commencement and continued refinement of the project's machine learning framework. Initial predictive models were developed using the processed climate and malaria datasets. Baseline algorithms were implemented to evaluate predictive performance, identify influential environmental variables, and assess data suitability. Under my mentor's guidance, various machine learning approaches, including Random Forest, Gradient Boosting, Long Short-Term Memory (LSTM), and XGBoost, were explored to improve predictive accuracy and robustness. Community knowledge obtained through field engagement is also being incorporated into the modelling framework to complement quantitative environmental data and improve the contextual relevance of prediction outputs.

A significant milestone was a six-week research residency at the University of Pretoria, under the mentorship of Professor Vukosi Marivate, in the Data Science for Social Impact (DSFSI) Laboratory. During this period, the machine learning framework evolved from initial baseline models to a comprehensive model tournament that compared multiple predictive algorithms and identified the most robust approaches for district-level malaria prediction. Explainable artificial intelligence techniques, particularly SHAP (SHapley Additive Explanations), were integrated into the modelling pipeline to improve transparency and interpretability by identifying how environmental variables such as rainfall, humidity, and temperature contribute to localised malaria outbreaks. Access to high-performance computational resources during the visit also enabled efficient model training and evaluation, overcoming computational limitations experienced earlier in the project.

Community Engagement

Community engagement constituted another major accomplishment during the reporting period. Between 23 and 27 February 2026, comprehensive field activities were conducted within the Wa Municipality in Ghana. Engagements included consultations with regional and municipal health directorates, household surveys, key informant interviews with healthcare professionals, focus group discussions, and participatory mapping exercises across five communities representing urban, peri-urban, and rural settings. In total, 350 household surveys were completed, together with six key informant interviews and separate focus group discussions involving male and female community members. The engagement generated valuable local knowledge regarding perceived climate variability, mosquito ecology, malaria transmission, and adaptation practices. Both healthcare professionals and community members consistently reported noticeable changes in rainfall and temperature patterns that they believe are influencing mosquito breeding habitats and increasing malaria risk. These findings reinforce the importance of integrating local knowledge with machine learning models to support climate-informed malaria early warning systems.

One notable experience during the fieldwork in the Wa Municipality occurred during community engagement activities when initial expectations about participation patterns differed significantly from what was observed in practice. In some communities, what was planned as structured household surveys gradually evolved into open household discussions, as residents spontaneously joined and began contributing their own observations about climate variability and malaria trends. While this initially required adjustments to the facilitation approach and time management, it ultimately enriched the data collection process by allowing a broader range of voices to be heard than originally anticipated.

Another interesting experience emerged during interactions with healthcare professionals, where informal discussions often extended beyond the formal interview guides. In several instances, clinicians shared detailed historical accounts of malaria trends in the region and described how their clinical intuition had evolved alongside changing environmental conditions. These moments provided deeper contextual insights that complemented the structured data and strengthened the interpretation of epidemiological patterns.

Scientific Collaboration and Capacity Building

The research residency at the University of Pretoria substantially strengthened the project's collaborative dimension. Interactions with researchers from the University of Pretoria, the Institute for Sustainable Malaria Control, and the Lancet Countdown Africa expanded the research's interdisciplinary scope. They aligned the project with the One Health framework. These collaborations have created opportunities for future joint research, capacity building, and broader scientific engagement while enhancing my leadership and research skills through the Programme.

Research Dissemination

The project also contributed to scientific communication and knowledge dissemination during the reporting period. Research findings and methodological developments have been presented through scientific webinars. In contrast, research abstracts have been submitted to international conferences, including the Deep Learning Indaba and the Africa Data Science Conference. These dissemination activities have provided opportunities to gather feedback from international experts and to increase the visibility of African-led artificial intelligence applications for climate-sensitive disease surveillance.

Challenges Encountered and Mitigation Measures

Several challenges were encountered during the reporting period. Administrative delays affected the ethical clearance process, while personnel changes at collaborating institutions temporarily slowed stakeholder engagement and data acquisition. Sparse meteorological station coverage and missing climate observations required the integration of satellite-derived datasets to ensure continuous spatial coverage.

Limited computational resources and data imbalance across districts initially constrained the development of machine learning models. However, access to computational infrastructure during the University of Pretoria research visit substantially addressed these limitations. Community engagement activities also coincided with the fasting period within participating communities, requiring adjustments to field schedules, extended engagement periods, and modifications to daily working hours to maintain participant involvement and data quality.

Conclusion and Next Steps

Overall, the first year of the Programme has successfully established the scientific, institutional, and technical foundations for the project. Ethical approvals have been secured, strategic partnerships have been developed, extensive climate and health datasets have been assembled, meaningful community engagement has been completed, and advanced machine learning model development is well underway.

During the second year, the project will focus on refining and validating predictive machine learning models, integrating community-derived knowledge into the modelling framework, developing climate-informed malaria early warning tools, and translating research findings into decision-support products for public health practitioners and policymakers. The mentorship, collaborative opportunities, and capacity-building support provided through the FAR-LeaF Programme have been instrumental in achieving these milestones and continue to strengthen my research, leadership, and interdisciplinary collaboration skills.

Annual report submitted by Dr Jacob Ageykum

(summarised for publication by Heidi Sonnekus for the FAR-LeaF Programme)