Analysis of Accuracy in Predicting Stability of Piers Evaluated by Impact Load Test

Abstract

The safety of piers is predicted through machine learning based on a database built from impact load tests on piers. However, since the amount of data is insufficient to learn the algorithm using only field experiment data, the accuracy of the algorithm is analyzed by constructing 1,159 data sets of a model experiment and numerical analysis data. Among machine learning algorithms, the accuracy of safety diagnosis prediction is studied using three algorithms: Support Vector Machine (SVM), Decision Tree (DT), and Logistic Regression. The algorithm is evaluated using an evaluation index according to a confusion matrix and, as a result, the highest accuracy is shown at 95.9% when the support vector machine, a binary classification model that predicts the boundary line between the two classes of safety and poor, is used. © 2021 The Korean Society for Railway. All rights reserved.