Capturing the Value of Predictive Modeling in Logistics

Editor’s Note: Every year, 40 or so students in the MIT Center for Transportation & Logistics' (MIT CTL) Master of Supply Chain Management (SCM) program complete one-year thesis research projects. The students are early-career business professionals from multiple countries with 2 to 10 years of experience in the industry. Most of the research projects are chosen, sponsored by, and carried out in collaboration with multinational corporations. Joint teams that include MIT SCM students and MIT CTL faculty work on the real-world problems. In this series, we summarize a selection of the latest SCM research.

Third party logistics providers (3PLs) are constantly fighting for competitive advantage. In a fragmented industry, having an edge is critical to success. Our thesis research sponsor company, a leading 3PL, gains that edge through a proprietary software. Their software matches shippers with carriers in a way that facilitates very high service levels. Their interest now is to enhance their success with predictive analytics techniques.

We developed a logistic regression model for the company based on a historical data set it provided. The model indicates which loads are likely to be delayed. By tracking only loads dictated by the model, the company can achieve significant resource reduction.

Logistic Regression vs Other Models

Logistic regression was appealing for this research due to its robustness and intuitive nature. Managers can observe which combinations of variables are used to make the predictions. They can also leverage reasoning to explain any variance. These model properties differ from those of other popular modeling methods. For example, neural networks and random forests methods provide alternatives that are essentially “black box” approaches. The user does not interface with what is happening in the model.

Despite our ultimate selection of the logistic regression model, we used the other models for validation. In this way, we achieved a deeper understanding of what the data was telling us.

Valuable Insights

Though the model could be used purely for resource allocation, some interesting insights and questions arose from the research.

First, the statistical power of one particular combination of variables was extraordinary. Significant value could be extracted from focusing on those attributes.

Next, the research highlighted some interesting questions about how such a model might be applied. Can the predictions be employed to make a change in the planning process, thereby avoiding a delay? Such a proactive approach could boost service levels even higher.

Lastly, our model offered substantial predictive value. That is, the model was extremely accurate at classifying loads that were on time. Having a high degree of certainty that resources should not be wasted on those loads is a valuable output for 3PLs.

Predictive models progressing

In the past, a lack of data and inadequate processing power rendered predictive analytics infeasible. Times have changed. As technology and innovation drive forward, methods facilitating the use of mathematical tools for predictions improve. Data collection and computing power are two key areas where improvements are changing the landscape of predictive applications.

The Internet of Things (IoT) and Cloud storage have transformed data collection and storage capacity. Further, statistical software packages can crunch large datasets and provide companies with valuable operational insights. These packages are supported by significant improvements in computing power and availability.

Ultimately, advances in these areas will enable researchers to make accurate predictions on freight movement disruptions. Firms at the forefront of these modeling methods will gain competitive advantage.

The SCM thesis Predicting On-Time Delivery in the Trucking Industry was authored by Rafael Duarte Alcoba and Kenneth W. Ohlund and supervised by Dr. Matthias Winkenbach, Research Associate, MIT Center for Transportation & Logistics. For more information on the research please contact Matthias Winkenbach at [email protected].

SC
MR