A comparative evaluation of machine learning classification models utilizing GIS and key flood-causing factors for enhancing flood vulnerability assessment

Recent flood events in Jeddah have caused substantial damage and loss of life, underscoring the urgent need for accurate flood vulnerability assessments to support disaster risk reduction and enhance urban resilience. This work integrates machine learning (ML) techniques with Geographic Information System (GIS) data to classify flood-prone areas across the region. Six critical flood-causing factors—surface elevation, slope, rainfall, proximity to streams, topographic wetness index, and curve number—were derived from GIS layers to capture the topographic, hydrological, and land surface characteristics that contribute to flood generation. Five ML classifiers were tested for their predictive performance: Logistic Regression, Coarse Tree, Fine Tree, Quadratic Support Vector Machine (SVM), and Narrow Artificial Neural Network (ANN). In order to guarantee the models’ resilience and applicability, the dataset was separated into subsets for training (70%) testing (15%) and validation (15%). Area Under the Curve (AUC) metrics, parallel coordinate plots, confusion matrices, and Receiver Operating Characteristic (ROC) curves were employed to thoroughly evaluate the model’s performance. With training accuracy of 95.7%, testing accuracy of 95.3%, and an AUC of 0.96, the Fine Tree classifier outperformed the other models under evaluation, demonstrating its higher predictive ability. Coarse Tree and Quadratic SVM followed with AUC values of 0.93 and 0.91, respectively, while ANN and Logistic Regression recorded 0.88 and 0.84. These findings demonstrate that integrating ML classifiers with GIS data offers a powerful, interpretable, and practical approach for identifying flood-vulnerable regions, providing essential guidance for sustainable urban planning and effective flood mitigation strategies in arid regions.