This study investigates the stabilization of lateritic soil through partial replacement of cement with flue gas desulfurization (FGD) gypsum, aiming to enhance its engineering properties for pavement subgrade applications. Lateritic soils are known for their high plasticity and low strength, which limit their utility in infrastructure. To address these challenges, soil specimens were treated with varying cement contents (3%, 6%, 9%) and FGD gypsum additions (1%-6%). Laboratory tests were conducted to evaluate plasticity, compaction, permeability, unconfined compressive strength (UCS), California Bearing Ratio (CBR), and fatigue behaviour. The optimal mix 6% cement with 3% FGD gypsum demonstrated significant improvements: UCS increased by over 110% after 28 days, permeability reduced by 26%, and soaked CBR improved by 56% compared to untreated soil. Additionally, fatigue life showed remarkable enhancement under cyclic loading, indicating increased durability for high-traffic applications. To support predictive insights, machine learning models including Decision Tree, Random Forest, and Multi-Layer Perceptron (MLP) were trained on 168 data samples. The MLP and Random Forest models achieved high prediction accuracy (R2 approximate to 0.98), effectively capturing the non-linear interactions between mix proportions and UCS. SHAP (SHapley Additive exPlanations) analysis identified curing duration as the most influential factor affecting strength development. This integrated experimental-computational approach not only validates the feasibility of using FGD gypsum in sustainable soil stabilization but also demonstrates the effectiveness of machine learning in predicting key geotechnical parameters, reducing reliance on extensive laboratory testing and promoting data-driven pavement design.

The presence of excavations or cavities beneath the foundations of a building can have a significant impact on their stability and cause extensive damage. Traditional methods for calculating the bearing capacity and subsidence of foundations over cavities can be complex and time-consuming, particularly when dealing with conditions that vary. In such situations, machine learning (ML) and deep learning (DL) techniques provide effective alternatives. This study concentrates on constructing a prediction model based on the performance of ML and DL algorithms that can be applied in real -world settings. The efficacy of eight algorithms, including Regression Analysis, k -Nearest Neighbor, Decision Tree, Random Forest, Multivariate Regression Spline, Artificial Neural Network, and Deep Neural Network, was evaluated. Using a Python -assisted automation technique integrated with the PLAXIS 2D platform, a dataset containing 272 cases with eight input parameters and one target variable was generated. In general, the DL model performed better than the ML models, and all models, except the regression models, attained outstanding results with an R 2 greater than 0.90. These models can also be used as surrogate models in reliability analysis to evaluate failure risks and probabilities.

Climate warming has aggravated the occurrence of thaw settlement in permafrost region, but the associated risk has not been precisely assessed or understood. This study applied four machine learning models to explore and compare the spatial distribution of thaw settlement risk in the Wudaoliang-Tuotuohe region, Qinghai-Tibet Plateau, namely, naive Bayesian, k-nearest neighbor, logistic model tree and random forest models. A total of 853 thaw settlement locations and 12 conditioning factors were used to train and validate the above four models. The results indicated that random forest model performed best with the highest accuracy. The risk map produced by random forest model implied that about 76.55% of thaw settlements were located in very high-risk regions, which only occupied 6.85% of study area. The volume ice content, active layer thickness and thawing degree days were the main factors leading thaw settlement. By further comparing the performances between random forest model and other three models, the overestimated and underestimated risk regions (Beiluhe and Tuotuohe basins), and imbalanced conditioning factors (altitude and slope angle) were determined. In contrast with similar studies, this research performed better in model construction and accuracy. The results can help designers to implement precautionary measures in thaw settlement risk management.