Development of geospatial optimized machine learning classification model for blackfly habitat suitability prediction

Onchocerciasis, commonly referred to as river blindness, represents a significant public health burden among Nigeria's neglected tropical diseases (NTDs). This parasitic infection is transmitted through the bite of infected female blackflies belonging to the Simulium genus, which maintain strict ecological requirements for breeding in fast-flowing, well-oxygenated riverine environments. The association between vector ecology and aquatic habitats creates distinct spatial patterns of disease transmission that demand rigorous geospatial analysis. Recent epidemiological surveillance has documented the persistent endemic nature of onchocerciasis across multiple regions of Nigeria, with considerable heterogeneity in disease prevalence and intensity, a critical gap remains in our ability to systematically identify and predict high-risk areas beyond locations with established surveillance infrastructure. This therefore poses substantial challenges for resource allocation, intervention planning, and progress toward elimination targets set by the World Health Organization and national control programs. The integration of machine learning methodologies into ecological niche modeling offers a transformative approach to addressing this challenge. Traditional statistical techniques for species distribution modeling, while foundational, exhibit significant limitations when confronted with the complexities of real-world epidemiological data. This study applies an optimized machine learning framework to predict blackfly habitat suitability and across Nigeria, combining support vector machines with butterfly optimization algorithms to address the dual challenges of feature selection and class imbalance inherent in vector distribution datasets. We obtained blackfly occurrence data from the Expanded Special Project for Elimination of Neglected Tropical Diseases (ESPEN), a WHO initiative focused on NTD control in Africa, spanning 2013 to 2024. Environmental and climatic datasets such as soil, aspect, elevation, hill shade, slope, population, occurrence proximity to water, occurrence of water bodies, temperature and precipitation respectively were sourced from multiple repositories and processed to Nigeria's geographic extent. We employed a Support Vector Machine algorithm for model development, with Butterfly Optimization Algorithm handling feature selection and model optimization. The SVM-BOA model achieved 75.4% accuracy in predicting blackfly presence. Feature importance analysis identified population density, elevation, hillshade, slope, temperature seasonality (BIO4), and proximity to water bodies as the primary determinants of blackfly habitat suitability. Model performance was evaluated using confusion matrices, and spatial predictions were generated to map habitat suitability of blackflies across Nigeria. This optimized framework provides a robust tool for identifying endemic areas and guiding targeted intervention strategies in Nigeria's onchocerciasis elimination program.