| 摘要: | Due to the issue of aging and deterioration has affected 1,138 bridges in Taoyuan over the previous 20 years. It has also resulted in 5 major bridge collapses in Taiwan in the last 5 years. Furthermore, based on Taiwan Vehicular Bridge Management System, over 21% of bridges need maintenance or even overhaul to keep performing the function. This research presents a Decision Tree (DT), Support Vector Machine (SVM), Multilayer Perceptron (MLP) Algorithms and Random Forest (RF) models as an empirical study. The comparison of the accuracy level from different machine learning in this research predicts the urgency to be maintained (U) , which assisted to evaluate and manage existing bridges in Taoyuan city and investigated the applicability of Decision Tree (DT), Support Vector Machine (SVM), Multilayer Perceptron (MLP) and Random Forest (RF) model.;An increase in the worldwide accident rate often occurs in the construction industry. The difficulties that construction engineers face in implementing bridge construction include not only implementation methods, but also force majeure situations such as floods, corrosion deterioration and typhoons, particularly natural disasters that occur unexpectedly. The issue of aging and deterioration has affected 1,138 bridges in Taoyuan over the previous 20 years. It has also resulted in 5 major bridge collapses in Taiwan in the last 5 years. Furthermore, based on Taiwan Vehicular Bridge Management System, over 21% of bridges need maintenance or even overhaul to keep performing the function. This research presents a Decision Tree (DT), Support Vector Machine (SVM), Multilayer Perceptron (MLP) Algorithms and Random Forest (RF) models as an empirical study. The comparison of the accuracy level from different machine learning in this researchpredicts the urgency to be maintained (U) , which assisted to evaluate and manage existing bridges in Taoyuan city and investigated the applicability of Decision Tree (DT), Support Vector Machine (SVM), Multilayer Perceptron (MLP) and Random Forest (RF) model. The datasets collected 6,137 records regarding bridge maintenance for the entire bridges in Taoyuan region from April 2021 to March 2022 in Taiwan Vehicular Bridge Management System. The input space of the model is made up of six features, namely, age of bridge, length of the bridge, width of bridge, span of bridge, bridge component, damage location and quantity. The results show that the accuracy test using Random Forest (RF) model has the higher accuracy or correctly classified rate of xxx than using other 3 models. When it is specified to predict data bridge deterioration, the random forest is the most suitable method that enables the ensemble modelto involve training a large number of decision trees and to combine their predictions. The research findings have proposed several suggestions for further studies such as improving data collection method, collecting more sufficient data, improving data preprocessing, and analyzing and evaluating the most suitable algorithm, which may facilitate handling missing data obtained from random forest and decision tree model. |
