Date of Graduation

5-2026

Document Type

Thesis

Degree Name

Bachelor of Science in Data Science

Degree Level

Undergraduate

Department

Data Science

Advisor/Mentor

Dr. Karl Schubert

Committee Member

Dr. Michael Kochen

Second Committee Member

Dr. Md Amzad Hossain

Abstract

In the highly competitive truckload freight market, competitive pricing is essential. Modern transportation firms employ a multitude of pricing analysis tools to develop pricing decisions that balance profitability with the probability of winning customer bids. Using proprietary ArcBest shipment data from 2023 to 2025, this thesis develops a decision-support framework for strategic trucking brokerage pricing in the face of incomplete information. The analysis has two parts. First, regression models—including regularized linear models, random forests, and gradient boosted trees—are used to estimate the expected margin of a shipment conditional on customer, lane, and market characteristics. Second, due to the absence of observed losing bids, a one-class classification framework is introduced to approximate bid competitiveness by learning the distribution of historically accepted (winning) quotes and assigning anomaly scores as a proxy for competitiveness. Anomaly scores are a relative measure of typicalness of an observation; relatively atypical observations are assigned high anomaly scores, while relatively typical observations are assigned low anomaly scores. This modeling strategy is motivated by the incompleteness of the available pricing data. ArcBest observes detailed information on accepted shipments but does not observe a comparable set of rejected quotes or competing bids. Thus, the framework does not directly estimate customer acceptance probabilities but instead extracts pricing signals from the observed distribution of historical wins.

Gradient-boosted models performed best for margin prediction, with lane conditions, customer frequency, and market-rate ratios emerging as important predictors. The one-class models identify a plausible region of feature space occupied by historical wins; quotes with higher anomaly scores look less like those wins and are therefore treated as less competitive. Linking these components confirm a clear empirical tradeoff between margin and competitiveness, consistent with downward-sloping demand in margin space. Taken together, the models let analysts evaluate a quote along two dimensions: expected margin and relative competitiveness. That gives a more structured way to price quotes even without observed losses.

In addition, the study explores whether more aggressive winning bids are associated with less favorable realized shipment outcomes, providing preliminary evidence relevant to the Winner’s Curse hypothesis in freight brokerage settings. Multiple linear regressions were developed to explore this relationship. The regressions revealed counter-intuitive results: worse operational outcomes were associated with less aggressive winning bids. However, load difficulty was identified as a potential confounder. More difficult loads naturally have worse operational outcomes but are potentially subject to less aggressive bidding due to lower volume of bidding overall which would obfuscate the causal relationship between bid aggression and operational outcomes. Potential stronger causal identification strategies are named but left undeveloped.

Overall, the results suggest that win-only quote data still contain useful pricing information, even without complete bid histories. The central contribution of this thesis is showing how pricing decision support can be developed even when the bid environment is only partially expressed.

Keywords

Truckload brokerage; freight pricing; margin prediction; win-only data; one-class classification; anomaly detection; machine learning; gradient boosted trees; random forest; missing counterfactuals; pricing decision support; Winner’s Curse

Download

Available for download on Sunday, May 06, 2029

Share

COinS