Confidence-Aware Deep Learning for Load Plan Adjustments in the Parcel Service Industry

TL;DR

The findings suggest that data-driven methods can substantially improve decision making in inbound load planning, offering planners a comprehensive, trustworthy, and real-time framework to make decisions.

Abstract

This study develops a deep learning-based approach to automate inbound load plan adjustments for a large transportation and logistics company. It addresses a critical challenge for the efficient and resilient planning of E-commerce operations in presence of increasing uncertainties. The paper introduces an innovative data-driven approach to inbound load planning. Leveraging extensive historical data, the paper presents a two-stage decision-making process using deep learning and conformal prediction to provide scalable, accurate, and confidence-aware solutions. The first stage of the prediction is dedicated to tactical load-planning, while the second stage is dedicated to the operational planning, incorporating the latest available data to refine the decisions at the finest granularity. Extensive experiments compare traditional machine learning models and deep learning methods. They highlight the importance and effectiveness of the embedding layers for enhancing the performance of deep learning models. Furthermore, the results emphasize the efficacy of conformal prediction to provide confidence-aware prediction sets. The findings suggest that data-driven methods can substantially improve decision making in inbound load planning, offering planners a comprehensive, trustworthy, and real-time framework to make decisions. The initial deployment in the industry setting indicates a high accuracy of the proposed framework.

Figures (11)

• Figure 1: Illustration of an internal shift, where load $3$ was planned to be processed during sort $S1$ in building $E$ but was actually processed during sort $S2$ in building $E$.
• Figure 2: Illustration of an external shift, where load $3$ was planned to be processed during sort $S2$ in building $D$ but was actually processed during sort $S2$ in building $E$.
• Figure 3: Number of loads through different dimensions: (a) The number of loads per shift class. (b) The number of loads processed in different buildings. (c) The number of loads processed in each sort. (d) The number of loads processed during each day of the week.
• Figure 4: The Framework description. First, the data is embedded through dedicated layers for categorical and numerical features respectively. Then, a neural network predicts the building of the load. This prediction is used as an input feature for the sort prediction, which has a similar architecture. The sort prediction is refined with additional information on the day of operations (i.e., Stage 2). Conformal prediction is performed for both building and sort predictions to provide prediction sets with a pre-specified confidence level. The figure illustrates the cumulative probabilities with classes ordered in decreasing order of estimated probabilities.
• Figure 5: Categorical Embeddings: Value $x^{cat}_i = A$ of feature $x^{cat}$ for sample $i$ is embedded as vector $W_A$ using the Look-up table of the corresponding feature.