Supervisor

Taufique Ahmed

Programme

HDIP in Data Analytics for Business

Subject

Computer Science

Abstract

Product returns are a significant challenge for online retailers, contributing to additional operational costs and affecting inventory management and profitability. This project develops and evaluates three machine learning models to predict product returns for a UK online retailer, building on the data understanding and preparation completed in the previous CA2 capstone project. Following the CRISP-DM framework, the study extends the analysis through enhanced exploratory data analysis, feature engineering, model development, hyperparameter optimisation, cross-validation, and model interpretability. Logistic Regression, a Tuned Random Forest, and XGBoost are evaluated, with the Tuned Random Forest achieving the strongest overall performance, obtaining an F1 score of 0.69 and a ROC-AUC of 0.978. Lowering the classification threshold to 0.3 increased recall to 0.80, demonstrating the potential to identify a greater proportion of high-risk transactions. Model interpretability is addressed using Permutation Importance for global feature analysis and LIME for explaining individual predictions. The report also considers the strategic, legal, and ethical implications of deploying machine learning in online retail through SWOT, PESTLE, GDPR, EU AI Act, and Digital Services Act considerations. The findings demonstrate the potential of machine learning as a decision-support tool for identifying transactions with a higher likelihood of return and supporting more informed operational and business decisions.

Date of Award

2026

Full Publication Date

2026

Access Rights

open access

Document Type

Capstone Project

Resource Type

thesis

Download

Included in

Data Science Commons

Share

COinS