Machine learning models for predicting the residual value of heavy construction equipment: An evaluation of modified decision tree, LightGBM, and XGBoost regression

Abstract

It is challenging to develop accurate models for heavy construction equipment residual value prediction using conventional approaches. This article proposes three Machine Learning-based methods of Modified Decision Tree (MDT), LightGBM, and XGBoost regressions to predict construction equipment's residual value. Supervised machine learning algorithms were used to comprehensively investigate the datasets throughout training, testing, modeling, and cross-validation processes. Four performance metrics (i.e., MAE, MSE, MAPE, and R2) were used to measure and compare the algorithms' accuracy. Based on the coefficient of determination results, the MDT algorithm has the highest prediction accuracy of 0.9284, versus the LightGBM with an accuracy of 0.8765, followed by XGBoost, obtaining an accuracy of 0.8493. The MDT can be used as a managerial decision support tool for equipment sellers, buyers, and owners to perform equipment life cycle analysis and take equipment selling, purchasing, overhauling, repairing, disposing, and replacing decisions. Thus, this study motivates machine learning's potential to help advancing automation as a coherent field of research within the construction industry.