MACHINE LEARNING ALGORITHMS APPLIED TO THE PREDICTION OF SLOPE FAILURES IN EARTH DAMS TRIGGERED BY RAINFALL

The stability of slopes in earth dams is a matter of global significance. Precipitation is a crucial factor in this analysis, as rainfall is a major trigger for landslides. Recent advancements in computational system that integrate numerical models with machine learning algorithms offer a robust approach for accurate predictions. This investigation employs seven widely used algorithms for predicting slope stability: artificial neural networks, support vector machines, decision trees, random forests, k-nearest neighbors, logistic regression, and Naive Bayes. This research utilizes a dataset with 5637 cases of earth dams generated through coupled numerical methods, which consider transient seepage and the impact of precipitation on slope stability. The goal is to develop prediction models with high accuracy. The results indicate that the best overall response case is k-nearest neighbors with an accuracy of 0.983 and an error of 0.017; followed by artificial neural networks with an accuracy of 0.966 and an error of 0.034. The most unfavorable results were obtained for Naive Bayes with accuracy of 0.852 and an error of 0.148; followed by support vector machines with an accuracy of 0.863 and an error of 0.137. In addition, three real-worlds slopes with similar characteristics to those proposed in the dataset were selected to validate the individual results obtained with each of the seven algorithms. From this analysis it was detected that artificial neural network, support vector machines, and random forests were the only ones that accurately predicted the expected response in the validation.