JMIR Publications

ABSTRACT

Background:

In the era of rapid digital evolution, artificial intelligence (AI) and machine learning (ML) have emerged as valuable tools for organizations seeking to extract insights from large datasets. However, the adoption of AI and ML, traditionally associated with technical domains like computer science and data science, poses challenges for non-technical professionals such as public health researchers. While existing no-code tools offer potential solutions, they are rarely tailored to the specific research workflows and local contexts of African institutions.

Objective:

This study aimed to design, develop, evaluate, and deploy the AutoML No-Code platform that supports end-to-end predictive analytics across different domains, with demonstrated applications in public health, such as disease risk prediction, stroke outcome modelling, air pollution–mortality analysis, and maternal and child health research.

Methods:

The platform was co-developed using an iterative, Agile approach in collaboration with public health researchers at the African Population and Health Research Center (APHRC). Version 1.0.0 was informed by a comprehensive needs assessment involving 7 researchers. Version 1.0.0 was informed by a needs assessment involving seven researchers. Version 2.0.0 introduced significant enhancements, including Observational Medical Outcomes Partnership (OMOP-based) analytics, data anonymization tools, automated research question generation using a large language model, and deep learning workflows for image classification, segmentation, and object detection. Version 2.0.0 was evaluated across two structured workshop settings: the Machine Learning for Health (ML4H) Workshop in Entebbe, Uganda (November 2025), where 21 participants from six African institutions applied the platform to real-world public health datasets using a mixed-methods approach combining structured closed and open-ended questionnaires, and a two-week training at Kampala International University (KIU) under the Data Science Without Borders (DSWB) project, where 26 participants completed an end-of-training survey. The platform supports the entire analytics pipeline, covering data ingestion, preprocessing, model training, evaluation, and interpretation.

Results:

Version 2.0.0 was well received across both settings, with participants reporting improvements in usability and strong acceptability of the platform. At the ML4H Workshop, the platform was successfully applied to 14 diverse public health use cases, including modelling the relationship between air pollution and mortality, predicting stroke outcomes, and analysing maternal and child health, infectious disease, and mental health outcomes, with several analyses generating manuscript-ready outputs. At KIU, 24 out of 26 (92%) participants agreed or strongly agreed that the training improved their understanding of ML concepts, and 24 out of 26 (92%) reported enhanced skills in data management and visualization, despite 16 out of 26 (62%) having no prior machine learning experience. Users across both settings highlighted the platform's capacity to unify data preprocessing, model development, and evaluation within a single interface, reducing reliance on programming skills. However, evaluation was conducted across a relatively small number of institutions and participants, which may limit the generalizability of findings.

Conclusions:

The AutoML No-Code platform provides a flexible no-code environment for AI and ML-driven predictive analytics across domains, with particular relevance for public health research in Africa. Deployment across multiple institutions has provided early evidence of its practical utility in supporting researchers with limited programming backgrounds. Further evaluation across diverse user groups and larger-scale deployments is needed to better understand its generalizability and long-term impact on research productivity and decision-making.