Blind prediction of slope safety factors using machine learning_pre-train model

This package provides the supporting materials developed from the study on slope reliability analysis integrating the Random Finite Element Method (RFEM) and Convolutional Neural Networks (CNNs). The RFEM incorporates the concept of random fields to model the spatial variability of soil properties. The study developed a machine learning approach to efficiently predict slope safety factor by training CNN models on RFEM simulation results. The compressed file includes:Pre-trained CNN model – developed using RFEM simulation outputs for slopes with 30° and 50° inclinations, accounting for soil spatial variability through the coefficient of variation and scale of fluctuation. The model has also been tested on unseen 40° slopes to assess its generalization capability. RFEM-based model – constructed to generate simulation results that serve as the training dataset for the CNN framework. Partial dataset – selected samples demonstrating the data structure and input–output relationships for training and prediction. Results from the study indicated that both shallow CNN and ResNet-enhanced architectures provided accurate predictions of slope safety factors, with the ResNet-6 configuration showing the best performance and generalization across different slope geometries. Please note that only part of the dataset is included in this release. Users who require the complete dataset or additional materials are encouraged to contact the original creators directly.