Smart Modeling Approaches for Foundation-Settlement Forecasting: A Comprehensive Review (2015–2025)

Foundation settlement is a critical concern in geotechnical engineering because it directly affects the stability, safety, and serviceability of civil infrastructure. Traditional empirical and numerical approaches, while useful, often fall short of capturing the highly nonlinear and heterogeneous behavior of soil–structure interaction. Over the past decade—particularly between 2015 and 2025—machine learning (ML) has emerged as a transformative tool for predicting foundation settlement, offering improved accuracy, adaptability, and computational efficiency. This chapter provides a systematic review of ML applications published from 2015 to 2025 across shallow foundations, deep foundations, highway embankments, and dam structures. The findings indicate that hybrid and ensemble models—such as ANN–PSO, GA–ANN, and CNN–LSTM—consistently outperform standalone algorithms in predictive accuracy and robustness. Time-series models, particularly LSTM networks, show strong performance when the temporal evolution of settlement is critical. Nevertheless, challenges remain, including limited dataset sizes, the absence of standardized repositories, model interpretability, and difficulties in transferring trained models across diverse geotechnical conditions. Beyond summarizing progress during 2015–2025, the chapter identifies key research gaps and emerging directions: integrating ML with IoT-based monitoring systems; developing explainable and probabilistic ML for risk-informed design; and adopting physics-informed learning frameworks that embed geotechnical principles within data-driven models. By critically analyzing a decade of advances, this chapter underscores a paradigm shift from conventional settlement prediction to intelligent, hybrid, and interdisciplinary approaches—paving the way for more resilient and sustainable geotechnical infrastructure design.