Ph.D. Thesis Defense Announcement
Enhanced Construction Cost Estimation of Highway Projects using Emerging Statistical and Machine Learning Techniques
By
Mingshu Li
Advisor(s)
Dr. Baabak Ashuri (CEE/BC)
Committee Members:
Dr. Baabak Ashuri (CEE/BC), Dr. Patricia L. Mokhtarian: (CEE), Dr. Eric Marks (CEE), Dr. Polo Chau (CSE), Dr. Minsoo Baek (CM, KSU)
Date & Time: March 29th, 3:30 pm
Abstract
Several state departments of transportaon (state DOTs) have encountered significant
challenges in accurately esmang costs for their highway projects, oen resulng in
discrepancies between the states’ DOT esmates (owner’s esmates) and contractors’
submied bids. These inaccuracies can lead to cost overrun, scope change, schedule delay,
postponement, and cancellaon of transportaon projects, which are problemac for both
owner organizaons and highway contractors. There is a crical need to enhance the quality of
construcon cost esmates to efficiently allocate public funds and increase confidence in
engineer’s esmates. Addressing this need, the overarching objecve of this research is to
advance construcon cost esmaon for highway projects through the applicaon of emerging
stascal modeling and machine learning techniques, examining cost esmaon at varying
levels of granularity for a comprehensive analysis.
The study first adopts a temporal perspecve at the monthly level, invesgang risk factors that
affect the accuracy of the owner’s esmate. This level of analysis allows for the examinaon of
several variables represenng the local highway construcon market, overall construcon market, macroeconomic condions, and the energy market to idenfy leading indicators of the
rao of low bid to owner’s esmate. Appropriate me-series models, such as ARIMAX, will be
applied to forecast this rao using idenfied leading indicators. This macro-level analysis offers
foundaonal insights into market trends and economic factors influencing cost esmaons,
seng the stage for more detailed invesgaons.
Transioning to the project level, the research conducts survival analysis to assess the
relaonship between several potenal drivers and the likelihood of inaccurate cost esmaon.
By innovavely applying concepts and methods from survival analysis to construcon cost
esmaon, this part of the study explores the impact of project-specific, bidder-specific, and
external market characteriscs on esmaon accuracy. This project-level analysis provides
crical insights into the dynamics at play within individual projects, complemenng the broader
market perspecve obtained from the temporal analysis.
Finally, at the most granular pay item level, forecasng models for early-phase cost esmaon
of lump sum pay items (Traffic Control and Grading Complete) are developed using text-mining
and machine learning techniques. This approach involves retrieving project informaon
available at the early stages of project development through text analysis and examining various
machine learning algorithms with idenfied key predicve features to select the bestperforming
model. By focusing on specific pay items, this level of analysis directly addresses the
praccal needs of designers and cost esmators, offering precise tools for early cost esmaon
and further enriching the comprehensive understanding gained from the previous analyses.
This research contributes to the body of knowledge through: (1) developing appropriate
mulvariate me-series models (i.e., ARIMAX models) to predict the rao of low bid to owner’s
esmate; (2) creang a Cox proporonal hazards model to explain and predict the likelihood of
inaccurate cost esmates; (3) developing machine learning algorithms to accurately esmate
prices of lump sum pay item at early stages of project development. It is ancipated that the
research outcome would help cost esmang professionals in transportaon agencies beer
understand the risk factors and potenal drivers of the deviaon between owner’s esmate and
low bids, prepare more accurate cost esmates and develop appropriate risk management
strategies for enhanced decision-making. Through its mul-level analysis, the study provides
significant insights into project planning, budget allocaon, and construcon cost management,
thereby underscoring the crical role of integrang machine learning and stascal modeling
techniques in enhancing the accuracy and reliability of cost esmaons for highway projects.