Abd. Rasyid Syamsuri, Rifki Arohman, Muhammad Renaldy Saputra, Muhammad Ikhlash, Sri Karyani Damanik

Integration of machine learning in e-commerce

Introduction

Integration of machine learning in e-commerce. Explore machine learning integration in e-commerce. This review analyzes ML for consumer behavior prediction & product recommendations, comparing algorithm performance & identifying research gaps.

Abstract

This systematic literature review examines the integration of machine learning (ML) in e-commerce, focusing on consumer behavior prediction and product recommendation systems. Following PRISMA guidelines, we searched Scopus, Web of Science, IEEE Xplore, and ACM Digital Library, identifying 1,247 records. After screening, 48 peer-reviewed articles (2019-2024) were included. This review makes three novel contributions: (1) a taxonomy of ML algorithms categorizing approaches by function (prediction vs. recommendation) and technique (supervised, unsupervised, deep learning); (2) a comparative analysis of algorithm performance across different e-commerce contexts; and (3) identification of specific research gaps requiring investigation. Findings reveal that hybrid recommendation systems combining collaborative filtering with deep learning achieve superior accuracy (mean improvement of 15-23% over single-method approaches), while gradient boosting methods (XGBoost, LightGBM) demonstrate the highest predictive performance for purchase behavior. Critical challenges include cold-start problems, data sparsity, algorithmic bias, and privacy concerns. We propose an integrative framework mapping ML technique to specific e-commerce applications and identify five priority areas for future research. Limitations include English-language restrictions and potential publication bias toward positive results.

Review

This systematic literature review provides a timely and rigorous examination of machine learning (ML) integration within e-commerce, specifically focusing on consumer behavior prediction and product recommendation systems. The authors meticulously adhere to PRISMA guidelines, employing a robust search strategy across major academic databases to identify a pertinent set of 48 peer-reviewed articles from 2019-2024. The paper's strength lies in its promised novel contributions: a well-structured taxonomy of ML algorithms by function and technique, a comparative analysis of their performance in varied e-commerce contexts, and the explicit identification of critical research gaps. This methodological rigor and clear framing establish a strong foundation for the review's subsequent insights. The findings presented are particularly insightful and hold significant practical implications for both researchers and industry practitioners. The observation that hybrid recommendation systems, combining collaborative filtering with deep learning, achieve substantial accuracy improvements (15-23% over single-method approaches) provides clear guidance for developing more effective recommendation engines. Similarly, the identification of gradient boosting methods (XGBoost, LightGBM) as superior for purchase behavior prediction offers valuable direction for predictive modeling. Beyond performance metrics, the review thoughtfully addresses the critical challenges faced in real-world ML deployment, including cold-start problems, data sparsity, algorithmic bias, and privacy concerns, underscoring a comprehensive understanding of the domain. The proposed integrative framework, mapping ML techniques to specific e-commerce applications, appears to be a valuable tool for future development, while the identification of five priority areas offers a strategic roadmap for advancing the field. While the authors acknowledge limitations such as English-language restrictions and potential publication bias, these are common to systematic reviews and do not detract significantly from the overall quality and depth of the analysis. This review is a highly commendable contribution, offering a structured understanding of the current landscape of ML in e-commerce and effectively charting a course for future research and innovation.

Full Text

Comments

You need to be logged in to post a comment.