Prediction of safety factors on slopes by using machine learning multilayer perceptron and decision tree techniques

Abstract

A more accurate strategy for measuring and predicting slope safety and risk was the study's goal. This study used machine learning methods to create accurate forecast models for the earth slope factor of safety (F.S). Multilayer perceptron (MLP) and decision tree (DT) have been chosen for the analysis. These methods have been widely used in geotechnical analysis and domain stability because of their unique benefits. We chose MLP, DT, and learning algorithmsas our core analytical approaches. The MLP and DT methods predicted the security component value as the main result of the earth slope mathematical models. In this study, 69.5 percent of the 105-group database was used to train the model, and 30.5% tested its accuracy. The controlled machine learning method-based decision tree (DT) and multilayer perceptron (MLP) computer models are used to analyze dump slope stability. SoftMax activates the output layer. The model summary showed 19.21 training cross-entropy errors and 11% erroneous predictions. The testing model had a 10.59 cross-entropy error and 21.9 wrong predictions. In the training classification phase, 36 viruses failed and two remained stable. This yields 94.7% accuracy. Testing classification yielded 18 failing viruses and 4 stable viruses. Maximum tree depth is 3, minimum parent node cases are 100, and minimum child nodes are 50. MLP modelshave superior accuracy, precision, and recall than DT models, which have 0.676, 0.604, and 0.711. The results summary reveals 3 nodes, 2 terminal nodes, and 1 depth. Therefore, MLP models outperform DT models.