Ehsan Foroutan

Extreme heat poses significant public health risks, particularly in rapidly urbanizing cities, and creates broad challenges for society. Traditional statistical and machine learning models are widely used to estimate heat‐related health outcomes, but their limited interpretability, high data requirements, and difficulty in incorporating expert knowledge, especially in data‐scarce settings, create key challenges for heat‐health modeling. To advance more sustainable and decision‐ready analytics, this study develops a hybrid fuzzy logic model and compares it with data‐driven and expert knowledge‐driven fuzzy logic–based approaches for estimating heat‐related emergency department visits (EDVs) in two major Texas metropolitan areas: Austin and Dallas–Fort Worth (DFW). The data‐driven model learns fuzzy rules, tunes membership functions, and optimizes rule weights from input–output data; the expert knowledge‐driven model uses domain knowledge to define rule structures; and the hybrid model combines expert rules with data‐driven tuning and optimization. All fuzzy models were evaluated with five‐fold cross‐validation and benchmarked against the best‐performing machine learning model identified in the AutoML framework—FLAML. Results show that in Austin, the hybrid fuzzy model achieved the lowest MAE of 118.73, outperforming the FLAML benchmark (MAE = 125.16), while the best‐performing data‐driven fuzzy model reached an MAE of 89.49. In contrast, in DFW, FLAML delivered the highest accuracy (MAE = 137.90), surpassing all fuzzy configurations. By systematically comparing expert‐driven, data‐driven, and hybrid fuzzy models against a modern AutoML benchmark, this study offers an original contribution that clarifies when and how interpretable, expert‐informed fuzzy systems can complement or exceed black‐box machine learning approaches, and enable agencies to make transparent, equitable, and uncertainty‐aware decisions that strengthen climate resilience and protect society.

This article presents a new methodology for urban growth modeling by integrating cellular automata (CA) and genetic fuzzy algorithms. The urban growth phenomenon is a spatio‐temporal and continuous process that can be modeled by employing fuzzy logic as a framework for handling vagueness. Dependence upon an expert for assigning membership functions and rule weights has made the fuzzy logic approach rather subjective. Our proposed methodology overcomes this shortcoming by applying genetic algorithms to tune and optimize the fuzzy logic system and to make it expert independent. The optimization process encompasses a fuzzy membership function and weights for fuzzy rules. This model uses linguistic variables for defining CA transition rules and applies them to represent the non‐deterministic nature of urban growth. The proposed model is applied to simulate urban growth in the Tehran Metropolitan Area in Iran across time steps of 1988, 1999, and 2010 developed using Landsat TM and ETM⁺ images and a Digital Elevation Model. The first data pairs were employed for calibration of the model parameters, and the remainder of the data was used for validation of the model across time. The model was evaluated using a relative operating characteristic of 0.88 and a figure of merit statistic of 0.31, which quantifies the model goodness of fit.