Using moving-window cross-validation algorithm to optimize bid-based cost estimating data usage

Transportation agencies provide guidance on using bid data from previous projects to prepare construction cost estimates that are based on assumptions whose validity is not challenged in the existing literature. Some of these assumptions refer to the optimal number of years of past data that should be used in bid-based estimating. This study aims to develop an approach that assists with the definition of optimal look-back periods for data retrieval to maximize cost estimating effectiveness in transportation construction projects. Identifying the optimal amount of data is irrelevant if it is not properly collected and processed. The proposed look-back period determination process is presented as part of a broader cost estimating system intended to ensure the effective use of the retrieved data. The paper presents an assessment of 20 cost indexing alternatives to identify the one that facilitates the highest estimating accuracy. The methodology consists of an innovative moving-window cross-validation algorithm designed to measure the performance of the estimating system for different look-back periods ranging from one to five years.