Given the presence of tourists in tourist‑oriented areas, the occurrence of environmental hazards in such areas can result in greater environmental, human, and economic damages and losses compared to other regions. Therefore, the aim of the present applied study is to delineate flood risk zones and identify villages exposed to flood hazards. This research was conducted as a case study in the Bisotun Tourist Sample Area, encompassing 60 villages. To achieve the main objective of the study, 13 criteria and two machine learning algorithms Classification and Regression Tree (CART) and Simple Random Forest (Simple RF) were employed. The results of the CART algorithm indicated that 39.5% of the study area and 80% of the villages are located within high and very high flood‑risk zones. The results of the Simple RF algorithm showed that 36.1% of the area and 75% of the villages fall within high and very high flood risk zones. A comparison of the flood risk zoning maps reveals a high degree of similarity between the results; however, the CART algorithm demonstrated a relatively more stringent performance in identifying high‑risk areas. The model validation results indicate that despite the high accuracy of both algorithms, the Simple RF algorithm achieved higher accuracy, with an AUC value of 0.985, compared to the CART algorithm with an AUC value of 0.942. Overall, the results of the study indicate that most tourist‑oriented villages are located within high‑risk flood zones.
Introduction
A review of the relevant literature indicates that artificial intelligence (AI) was first introduced by John McCarthy in 1955 as “the science and engineering of making intelligent machines.” Artificial intelligence generally refers to technologies derived from human cognitive processes that are implemented through computer systems. The expansion of intelligence as a theoretical concept emerged alongside efforts to simplify logical problem solving in computer science. In its early stages, artificial intelligence primarily supported theoretical reasoning and was mainly applied to solving complex mathematical problems. To date, no single comprehensive definition of artificial intelligence has been universally accepted, and various interpretations have been proposed. Most definitions conceptualize AI either as a subfield of computer science or in terms of machines’ ability to imitate human intelligence. Given its close association with the rapid development of computer technology, some researchers define artificial intelligence as “the scientific study of computational principles rooted in intelligent human thinking and behavior.” In general, artificial intelligence is regarded as a branch of advanced technologies capable of identifying, analyzing, and learning from the characteristics of human intelligence, thereby exhibiting intelligent behavior. One of the key applications of artificial intelligence in tourism development is the use of machine learning algorithms, which are increasingly employed to address tourists’ needs, such as multilingual communication and responses to complex inquiries. Tourism destinations particularly rural tourism sample areas face a variety of challenges, and machine learning techniques provide effective tools for managing these challenges. Among the most critical challenges confronting rural tourism areas is the occurrence of environmental hazards, especially floods, which can result in multidimensional consequences, including human, economic, environmental, and psychological impacts on tourists and local communities.
Material and Methods
The present study was conducted with the general objective of applying artificial intelligence in environmental risk management, with particular emphasis on flood risk in the Bisotun Tourist Sample Area. Given the complex and multi‑causal nature of flood events, an integrated framework combining field surveys, Geographic Information Systems (GIS), remote sensing data, and machine learning algorithms was adopted.
The integration of diverse spatial datasets with powerful decision‑making models—specifically the Classification and Regression Tree (CART) and Simple Random Forest (Simple RF) algorithms—enabled accurate identification of flood‑prone areas and analysis of the factors influencing flood occurrence. Overall, the research was carried out in seven main stages:

Literature review and research background

Previous national and international studies were reviewed to improve understanding of flood mechanisms and contributing factors.

Selection and definition of effective parameters

Based on literature review and expert opinions from the fields of geography, natural resources, environmental sciences, remote sensing, and crisis management, 13 key variables were selected as model inputs.

Flood occurrence data preparation and sampling

Sentinel‑1 radar imagery was used to generate flood occurrence point data, covering three critical periods: before, during, and after the 2019 flood event.

Data processing and spatial layer generation

Spatial datasets were collected and processed from multiple sources.

Implementation of machine learning models

The CART and Simple Random Forest algorithms were applied to model flood susceptibility.

Model performance evaluation

Model performance was assessed using the Receiver Operating Characteristic (ROC) curve and the Area Under the Curve (AUC).

Final map production and zoning analysis

Flood risk maps generated by the CART and Simple RF models were classified and integrated in the ArcGIS Pro environment.
Results and Discussion
The results of the CART algorithm indicated that 39.5% of the total study area and 80% of the 60 investigated villages are located within high and very high flood risk zones. Similarly, the Simple RF algorithm classified 36.1% of the study area and 75% of the villages as belonging to high and very high flood risk zones. Comparison of the flood risk zoning maps revealed a high degree of spatial similarity between the two models. However, the CART algorithm exhibited a relatively more stringent performance in identifying high risk areas. Model validation results showed that despite the high accuracy of both algorithms, the Simple RF model achieved superior predictive performance, with an AUC value of 0.985, compared to 0.942 for the CART model.
Conclusion
Overall, the outputs of machine learning algorithms demonstrate a high level of reliability in environmental risk management. In the context of flood risk management in the Bisotun Tourist Sample Area and its rural settlements, the main practical recommendations include: strict enforcement of construction regulations in flood‑prone areas; equipping rural settlements with modern flood early‑warning systems; ensuring timely and accurate flood forecasts by the Meteorological Organization; construction and development of flood diversion channels; use of durable and resilient materials in residential, public, and transportation infrastructure; dissemination of flood risk zoning maps among rural managers; establishment of relief and rescue centers near high‑risk villages; education of local communities on flood preparedness and response; and installation of warning signs in hazardous zones to inform both residents and tourists.