Opportunities and pitfalls of regression algorithms for predicting the residual value of heavy equipment — A comparative analysis

Abstract

The residual value of heavy equipment is essential for financial and economic considerations in the construction industry. In practice, empirical methods are frequently used to determine the residual value of a given piece of equipment. Here, various regression methods are compared based on a real-world dataset of used heavy equipment sales from a construction company. The results show that the prediction performance of traditional methods is clearly worse when compared to machine learning models not yet employed for this purpose. For the latter, preprocessing and parameter tuning are essential, and the article guides through these steps. Further, the article demonstrates how a variable importance value comparable across all methods can be obtained. These findings may also be useful in other applications.