Classification of soil layers in Deep Cement Mixing using optimized random forest integrated with AB-SMOTE for imbalance data

Deep Cement Mixing (DCM) technology is extensively used in geotechnical engineering to improve ground conditions, especially in projects involving soft soil foundations. Accurate classification of soil layers is essential to optimize DCM performance, ensuring infrastructure stability and safety. However, imbalances in soil layer data often lead to the underrepresentation of certain soil types, subsequently lowering classification accuracy and compromising model reliability. To address this issue, this paper presents an Adaptive Boundary-SMOTE (AB-SMOTE) algorithm that improves traditional oversampling methods by adaptively expanding minority class regions and identifying clean subregions to generate synthetic samples. Additionally, a Particle Swarm Optimization-Enhanced Random Forest (PSO-RF) model is developed to enhance soil layer classification, providing improved generalization and resilience against overfitting. Simulation results indicate that the AB-SMOTE algorithm effectively mitigates dataset imbalances by generating high-quality synthetic samples that significantly improve model training. Furthermore, applying the PSO-RF model to the balanced dataset yields notable improvements in classification accuracy, precision, and recall.