Predicting on-time deliveries in e-commerce: a machine learning approach to shipment performance analysis

Abstract

Timely delivery is a critical driver of customer satisfaction and loyalty in the competitive e-commerce sector. This study applies machine learning techniques to predict shipment punctuality using a dataset of 10,999 delivery records with 12 features covering customer behavior, product characteristics, and logistics details. The target variable, “Reached on Time,” indicates whether shipments were delivered on time or delayed. Through exploratory data analysis, the study identifies key delivery factors, including the influence of shipment weight, product cost, and discounts. Four supervised machine learning models—Logistic Regression, Decision Tree, Random Forest, and Support Vector Machine—are developed and evaluated using accuracy, F1-score, and AUC-ROC metrics. The results show that ensemble methods, particularly Random Forest, outperform traditional models in predicting late deliveries. The findings highlight the value of data-driven approaches for improving shipment reliability and provide actionable insights for optimizing logistics strategies and enhancing customer satisfaction.