Abstract
Background: Brucellosis (Malta fever) is a zoonotic bacterial disease prevalent in various regions of Iran, including Sabzevar County. This study aimed to model and predict the spatial incidence of human brucellosis using Geographic Information Systems (GIS) and machine learning algorithms, including Principal Component Analysis-Multilayer Perceptron (PCA-MLP) and Principal Component Analysis-Support Vector Machine (PCA-SVM).
Methods: This cross-sectional analytical study employed a computational modeling approach conducted in 2023 using data from 2011 to 2022. The data included epidemiological, demographic, climatic, vegetation, and topographic information. After processing in a GIS environment, the data were analyzed using PCA-enhanced MLP and SVM algorithms.
Findings: Seasonal pattern analysis revealed the highest disease incidence during spring in the years 2012, 2020, and 2022, as well as during the period from 2014 to 2017. In 2019 and 2021, the peak incidence shifted to summer. "Air humidity" and "altitude above sea level" were identified as the strongest predictive factors. The PCA-MLP model identified "Sheshtamad" and "Dastooran" as the main disease hotspots, while the PCA-SVM model detected a broader range of high-risk areas.
Conclusion: The PCA-SVM algorithm demonstrated superior accuracy in predicting the spatial pattern of brucellosis due to its enhanced capability to handle complex data. This model serves as an effective tool for targeted disease control planning in endemic areas.