Data Mining Algorithms for Classification Model of Engineering Grade

Abstract

This study uses various construction inspection data of the Public Construction Management Information System (PCMIS) from 2003 to 2018 and data mining (DM) to analyze the relationship among the engineering-grades, contract prices, project category, and progress of 1,015 cases and 499 defects. Association rule mining was used to derive 11 the rules which altogether contain the four types of defects and project attributes. The algorithms analyze the importance value of the attributes to gain relatively important defects. Therefore, association rules and important defects can provide a useful reference to enable engineering management personnel with aid to understand the relevance of defects and project attributes to improve construction quality. In addition, classification models are constructed using a neural network (NN), support vector machine (SVM), and C5.0 for DM algorithms. According to the results of accuracy and the area under the curve (AUC) show that the SVM has the best classification benefit, followed by NN. The classification model obtained herein can effectively predict the engineering-grade.