Backlog limitation system in Commercial Carrier hubs : Leveraging statistics techniques to limit the amount of backlog and late deliveries

Abstract: The main goal of Commercial Carriers is to deliver packages respecting the promises made to customers. Therefore, backlog prevention systems play a crucial role considering that backlogs would prevent CCs to reach their primary objective. This work aims at analyzing the state of the art of current available solutions to overcome these limitations, and, on top of those, a new approach is proposed. Then, the new proposal is evaluated and compared to already deployed solutions. The mechanism analyzed in this thesis aims at reaching, through the use of statistics, comparable results to ML solutions in terms of accuracy, but requiring less computational resources. To achieve this, the proposed methodology relies on a statistical model that does not need to be trained and executed, as typical ML models. As a consequence, one of the main benefits is the speed of execution: requiring less than 30 seconds, about 25 times faster than ML approaches. Our findings show that our approach requires about 15% of the data other solutions need: this is a significant advancement with respect to the current state of the art. This constitutes an additional benefit in terms of cost savings for tech resources, both in terms of data storage and program execution. Results prove the effectiveness of the proposal, with 80% accuracy, about 10% lower than averages ML approaches. Key point is the difference in the approach, for which is worth to evaluate the trade-off. Considering the effective low frequency of events that negatively impact the performances and result in backlog, the proposed approach would be valid for everyday scenarios but it would be still advisable to rely on even more accurate technologies during particularly intense periods.