دانشگاه اصفهان برنامه ریزی فضایی 2228-7485 16 3 2026 09 23 Modeling and Spatial Mapping of Cutaneous Leishmaniasis (CL) Distribution in Isfahan Province Using Machine Learning Algorithms مدل‌سازی و پهنه‌بندی توزیع مکانی بیماری سالک در استان اصفهان با استفاده از الگوریتم‌های یادگیری ماشین 29 50 30595 10.22108/sppl.2026.148536.1880 FA فاطمه عباسی دانش‌آموخته کارشناسی ارشد سنجش از دور و GIS، دانشکده علوم جغرافیایی و برنامه‌ریزی، دانشگاه اصفهان، اصفهان، ایران رضا ذاکری نژاد استادیار گروه جغرافیای طبیعی، دانشکده علوم جغرافیایی و برنامه‌ریزی، دانشگاه اصفهان، اصفهان، ایران عبدالله سیف دانشیار گروه جغرافیای طبیعی، دانشکده علوم جغرافیایی و برنامه‌ریزی، دانشگاه اصفهان، اصفهان، ایران Journal Article 2026 04 25 Abstract Cutaneous Leishmaniasis (CL) is a significant vector-borne parasitic disease with substantial public health, social, and economic impacts, particularly in endemic regions like Iran. This retrospective study aimed to model the spatial susceptibility of CL in Isfahan Province, Iran, from 2021 to 2023, utilizing 6,353 confirmed cases reported by local health centers. A comprehensive set of independent variables—including climatic indicators, topographic factors, vegetation indices (NDVI, EVI), land use, population density, and distance from water bodies—were integrated and analyzed within a Geographic Information Systems (GIS) framework. Spatial modeling was performed using LightGBM and Classification and Regression Trees (CART) algorithms with model performance evaluated through overall accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC). The LightGBM model demonstrated superior predictive performance (AUC=0.992, accuracy=96.55%) compared to the CART model (AUC=0.975, accuracy=96.65%). Spatial susceptibility mapping identified central and northwestern regions—including Isfahan, Khomeynishahr, Najafabad, Falavarjan, Barkhar, Mobarakeh, Kashan, and Aran and Bidgol—as high-risk areas. Variable importance analysis revealed that land use, seasonal average wind speed, distance from water bodies, and seasonal Enhanced Vegetation Index (EVI) were the most influential environmental predictors. These findings underscore the critical role of climatic and environmental factors in CL distribution and provide actionable spatial insights for targeted public health planning and vector control interventions in Isfahan Province. LightGBM is recommended as the preferred model for future susceptibility mapping endeavors. Keywords: Modeling, Cutaneous Leishmaniasis, Geographic Factors, Data Mining. Introduction Cutaneous Leishmaniasis (CL) is one of the most important vector-borne diseases, posing significant public health, social, and economic challenges worldwide. It is a parasitic disease that can manifest in cutaneous, mucocutaneous, or visceral forms (Shabanpour et al., 2022). CL is transmitted by female sandflies (Phlebotominae) and its causative agent is a protozoan parasite of the genus Leishmania. Transmission occurs from animal reservoirs, mainly domestic and wild rodents, to humans, causing skin ulcers that may persist for up to a year on the face, hands, and feet. These lesions impose psychological and social burdens, affecting patients' quality of life. In Iran, CL occurs in urban (dry) and rural (wet) forms. In urban areas, humans are the main reservoir, and Phlebotomus sergenti transmits Leishmania tropica. In rural areas, wild rodents act as reservoirs and Phlebotomus papatasi transmits Leishmania major (Mollalo et al., 2018; Shirzadi, 2012). Iran is considered an endemic focus with official records of approximately 20,000 new cases annually although actual numbers are likely higher (Shirzadi, 2012; Rathi & Hanafi-Bojd, 2006). Geographic factors play a critical role in disease distribution and Geographic Information Systems (GIS) provide tools for spatial analysis and visualization of CL patterns, facilitating public health planning and policy decisions (Bayatani & Sadeghi, 2012). Materials & Methods This retrospective study was conducted in Isfahan Province, Iran, from 2021 to 2023, by using all confirmed CL cases reported by health centers of Isfahan Province and Kashan County, totaling 6,353 records. The dependent variable was CL occurrence, while independent variables included climatic indicators (minimum, maximum, and mean temperature, precipitation, relative humidity, soil moisture, solar radiation, and wind speed), topographic factors (elevation, slope, and aspect), vegetation indices (NDVI and EVI), land use, population density, and distance from water bodies. Data were preprocessed, integrated, and analyzed in ArcGIS 10.8.2. Spatial modeling was performed by using LightGBM and CART algorithms in RStudio 2025.06.14 with 70% of the data used for training and 30% for testing. In the LightGBM model, parameters were set as follows: learning rate=0.05, number of trees (nrounds)=600, num_leaves=31, max_depth=−1, min_data_in_leaf=10, feature_fraction=0.8, bagging_fraction=0.8, and bagging_freq=5. In the CART model, the complexity parameter (cp) was set to 0.001, minsplit=20, minbucket=7, and maxdepth=30. Model performance was evaluated by using overall accuracy, sensitivity, specificity, ROC curves, and the area under the curve (AUC). High-risk areas were mapped and classified into 5 susceptibility classes, ranging from very low to very high and the relative importance of environmental variables was assessed for each model. Research Findings The spatial modeling of CL in Isfahan Province yielded distinct performance outcomes for the two algorithms employed. The LightGBM model demonstrated an overall accuracy of 96.55%, a Kappa coefficient of 0.9309, sensitivity of 95.44%, specificity of 97.59%, and an area under the curve (AUC) of 0.992. In comparison, the CART model achieved an overall accuracy of 96.65%, a Kappa coefficient of 0.9142, sensitivity of 97.61%, specificity of 94.00%, and an AUC of 0.975. Spatial susceptibility mapping based on the LightGBM-generated sensitivity map categorized Isfahan Province into 5 risk classes ranging from very low to very high. Peripheral areas, particularly eastern counties, such as Naein and Khur and Biabanak, exhibited very low susceptibility. In contrast, central and some northwestern regions demonstrated higher susceptibility to CL. Variable importance analysis revealed that the most influential environmental predictors in the LightGBM model were land use, seasonal average wind speed, distance from water bodies, and seasonal Enhanced Vegetation Index (EVI) values. The CART model showed consistent spatial patterns, identifying land use, wind speed, summer relative humidity, and EVI as key predictors. Overall, the findings confirmed that climatic and environmental factors—including land use, 3-year average EVI, relative humidity, and wind speed—were key determinants of CL distribution in the study area with high-risk areas primarily concentrated in central and northwestern regions, including Isfahan, Khomeynishahr, Najafabad, Falavarjan, Barkhar, Mobarakeh, Kashan, and Aran and Bidgol. Discussion of Results & Conclusion The performance evaluation of the LightGBM model indicated an overall accuracy of 96.55%, a Kappa coefficient of 0.9309, sensitivity of 95.44%, specificity of 97.59%, and an AUC of 0.992. The CART model achieved an overall accuracy of 96.65%, a Kappa coefficient of 0.9142, sensitivity of 97.61%, specificity of 94.00%, and an AUC of 0.975. The sensitivity map generated by LightGBM categorized Isfahan Province into 5 risk classes. Peripheral areas, especially eastern counties, such as Naein and Khur and Biabanak, exhibited very low sensitivity, whereas central and some northwestern regions showed higher susceptibility. Variable importance analysis revealed that the most influential factors in LightGBM included land use, seasonal average wind speed, distance from water bodies, and seasonal EVI values. Similarly, the CART model showed consistent spatial patterns with land use, wind speed, summer relative humidity, and EVI as key predictors. The performance evaluation showed that the LightGBM model (AUC=0.992) outperformed the CART model (AUC=0.975) in discriminating between infected and healthy areas. Therefore, LightGBM is recommended as the superior model for spatial susceptibility mapping of cutaneous leishmaniasis in Isfahan Province. CART, due to its simpler structure and interpretability, can serve as a baseline model for comparing other machine learning algorithms. Spatial sensitivity mapping revealed that high-risk areas were primarily concentrated in central and northwestern regions, including Isfahan, Khomeynishahr, Najafabad, Falavarjan, Barkhar, Mobarakeh, Kashan, and Aran and Bidgol. Modeling results confirmed that climatic and environmental factors, including land use, 3-year average Enhanced Vegetation Index (EVI), relative humidity, and wind speed, were key determinants of cutaneous leishmaniasis distribution. These findings are consistent with previous studies (Mollalo et al., 2018; Kosgei, 2024). چکیده لیشمانیوز پوستی از مهم‌ترین بیماری‌های منتقله از ناقل در ایران است و استان اصفهان به‌عنوان یکی از کانون‌های فعال آن، نیازمند شناسایی نواحی پرخطر جهت بهینه‌سازی برنامه‌های پیشگیری و کنترل است. هدف این پژوهش، مدل‌سازی و پهنه‌بندی حساسیت مکانی بروز لیشمانیوز پوستی در شهرستان‌های استان اصفهان طی سال‌های ۱۴۰۰ تا ۱۴۰۲ با استفاده از الگوریتم‌های یادگیری ماشین است. در این مطالعه گذشته‌نگر، متغیرهای اقلیمی، پوشش گیاهی، توپوگرافی، فاصله از آبراهه، کاربری اراضی و تراکم جمعیت به‌عنوان متغیرهای مستقل و موارد تأییدشده بیماری به‌عنوان متغیر وابسته در نظر گرفته شدند. داده‌ها پس از پیش‌پردازش در ArcGIS 10.8 با الگوریتم‌های LightGBM و CART در RStudio 2025.06.14 مدل‌سازی شدند. نتایج نشان داد مدل LightGBM با دقت 55/96%، ضریب کاپای 09/93% و سطح زیر منحتی (AUC) برابر 992/0 نسبت به مدل مدل CART با دقت 65/%96، ضریب کاپای 42/%91 و سطح زیر منحنی (AUC) برابر 975/0 عملکرد بهتری داشت. براساس مدل LightGBM به‌عنوان مدل برتر، پهنه‌بندی حساسیت مشخص کرد 61/%93 مساحت استان در کلاس بسیار کم‌خطر و تنها 43/%3 در کلاس‌های خطر زیاد و بسیار زیاد قرار دارد که عمدتاً در بخش‌های مرکزی و شمال‌غربی استان شامل شهرستان‌های اصفهان، خمینی‌شهر، نجف‌آباد، فلاورجان، برخوار، مبارکه، کاشان و آران و بیدگل متمرکز شده‌اند. مهم‌ترین عوامل مؤثر در پیش‌بینی بیماری شامل کاربری اراضی، سرعت باد، شاخص EVI و فاصله از آبراهه بودند. این یافته‌ها با شناسایی دقیق نواحی پرخطر، به برنامه‌ریزی هدفمند اقدامات پیشگیرانه و کنترلی در سطح استان کمک می‌کند.

https://sppl.ui.ac.ir/article_30595_9311de3328d81123ded8aa1ff6240591.pdf