Prediction of ocean import shipment lead time using machine learning methods

Abstract

This paper focuses on developing a predictive model for estimating the shipment lead time using machine learning methods for ocean import freight considering interests of different stakeholders such as shipper, carrier, freight forwarder, and consignee. Two different terminal criteria for calculating shipment lead time are defined considering different milestones: one with empty container return and the other with delivery confirmation at the destination. Real data obtained from an industry partner are used for implementation, and multinomial logistic regression is identified as the best classifier with the highest accuracy in each binning method, which is followed by a decision tree method. Additionally, commonly used classifiers such as multinomial logistic regression, decision tree, K-nearest neighbors, and support vector machine perform better than Naïve Bayes when the categorical variables are binarized, and vice versa when the categorical variables are converted into ordinal values. The proposed model has the significant potential to benefit different parties in the supply chain by providing improved visibility and predictability for shipment lead times.