In this study, the performance of modular flowerbox systems as semi-structural and bio-sustainable stabilizing elements was evaluated for improving the geotechnical behavior and stability of urban trenches in arid climates. To this end, six representative urban trenches were selected and subjected to field monitoring. Geotechnical data including natural moisture content, wet density, Standard Penetration Test (SPT) results, internal friction angle, effective cohesion, and recorded settlements over a six-month period were collected and analyzed. Stability analysis was conducted using finite element modeling in PLAXIS. In addition, data-driven approaches including XGBoost, Random Forest, and Multilayer Perceptron (MLP) neural networks were employed to predict settlement and analyze soil behavior. Model input parameters were calibrated based on laboratory results and instrumentation data, and model performance was evaluated using 10-fold cross-validation. Numerical modeling results indicated that installation of modular flowerboxes reduced surface settlement by 38% (from an average of 12.5 mm to 7.75 mm) and increased the factor of safety by 37% (from 1.15 to 1.57). Due to the shallow trench geometry, the influence of surface structural elements was clearly reflected in stress redistribution and displacement patterns. A reduction in horizontal stress ranging from 9 kPa to 13 kPa, along with decreased lateral wall displacement, was also observed. In the data-driven analysis, the XGBoost algorithm achieved the highest predictive accuracy, with an RMSE of 1.87 mm and a coefficient of determination (R²) of 0.94, demonstrating strong agreement with field instrumentation data. The findings suggest that integrating bio-sustainable surface systems with numerical modeling and intelligent data-driven techniques provides an efficient, cost-effective, and reliable approach for enhancing the stability of urban trenches, particularly in sandy–silty soils of arid regions.