Mapping the potential occurrence of Cutaneous Leishmaniasis using machine learning algorithms and remote sensing data

Cutaneous Leishmaniasis (CL) is a vector-borne parasitic disease closely linked to environmental conditions. This study aimed to model and map the spatial occurrence potential of CL across multiple study years in Golestan Province, Iran, using machine learning, ensemble learning approaches, and remote sensing data. A geospatial database was developed from CL case records aggregated at the level of cities and rural settlements during 2011-2013 and nine environmental variables, including precipitation, temperature, evapotranspiration, wind speed, drought conditions, vegetation cover, population density, elevation, and slope. Three machine learning algorithms-Random Forest (RF), Support Vector Regression (SVR), and Artificial Neural Network (ANN)-were applied. The outputs of the machine learning models were combined using majority voting and weighted voting techniques. Model performance was evaluated based on the Area Under the ROC Curve (AUC). Among individual models, ANN showed the highest performance (AUC = 0.843), while SVR had the lowest (AUC = 0.748). Ensemble approaches improved overall discriminatory performance, with weighted voting achieving the best overall performance (AUC = 0.911). Spatial results indicated that the northern and central areas had the highest CL occurrence potential, whereas the southern and southeastern regions had the lowest occurrence potential. These findings highlight the potential of integrating remote sensing data, machine learning algorithms, and ensemble methods for CL occurrence mapping.