EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

1573. Container dwell time prediction considering feature engineering and feature selection

Invited abstract in session WA-31: Analytics for Decision Making, stream Analytics.

Wednesday, 8:30-10:00
Room: 046 (building: 208)

Authors (first author is the speaker)

1. Mahdi Jahangard
Anglia Ruskin University
2. Ying Xie
School of Management, Cranfield University

Abstract

Container dwell time prediction plays a crucial role in optimizing logistics and supply chain operations at ports and container terminals. Accurate predictions of such an indicator can enhance resource allocation, reduce congestion, and improve overall efficiency at the point of the global trade network. Data preprocessing and feature selection are critical steps in developing robust and efficient predictive models. This study develops feature engineering not only for structured data but also for unstructured data where a textual feature is classified through GloVe word embeddings and Cosine similarity to leverage the strengths of both types of data into machine learning methods. In addition, various feature selection techniques such as Filter methods, Embedded methods, and Wrapper methods are applied to identify the most influential factors affecting container dwell time, ultimately selecting 12 out of the 19 features for prediction. Instead of developing a regression model to predict dwell time, we classify dwell time based on data distribution into two and three classes, highlighting the significance of dwell time ranges over singular values. Random Forest and XGBoost algorithms are employed for the classification, where the former outperforms in terms of accuracy. From an operational point of view, the results of classification can facilitate effective decision-making for operational management such as equipment scheduling, yard stacking planning, workload planning, etc.

Keywords

Status: accepted

Back to the list of papers