Data-driven classification of multi-source engineering waste soils for reuse purposes

: Rapid global urbanization and large-scale infrastructure projects, such as tunneling and piling, generate substantial Engineering Waste Soil (EWS). Conventional disposal methods are unsustainable, yet resource utilization is hindered by the lack of a suitable classification system. Conventional geotechnical standards, developed for natural soils, often fail to characterize the complex, multi-source, and chemically altered nature of EWS. This study proposes a data-driven rating framework to assess EWS reuse potential. An initial database of 554 samples was systematically compiled from international literature and field data. After strict screening, 296 valid samples were retained, and 43.23% missing values were addressed using Multiple Imputation by Chained Equations (MICE). Unsupervised K -means clustering identified three intrinsic data categories, bypassing a priori labeling biases and capturing the non-linear coupling of physicochemical properties. A Support Vector Machine (SVM) classifier trained on these labels achieved 98.89% accuracy. By synthesizing model predictions with 21 documented engineering cases and five major geotechnical standards (JTG, AASHTO, USCS, USDA, and JGS), a three-tier rating system was established: Type 1, Type 2, and Type 3. SHapley Additive exPlanations (SHAP) analysis confirmed the model's alignment with geotechnical mechanisms, identifying a robust granular skeleton (reflected by sand content) as the physical basis for the Type 1 designation, while the coupling of high plasticity index ( PI ) and clay content governs the classification of Type 3. This framework provides a scientific basis for the quantitative grading of waste soil.