LightPersML: A Lightweight Machine Learning Pipeline Architecture for Real-Time Personalization in Resource-Constrained E-commerce Businesses

Abstract

This paper presents a lightweight machine learning framework for e-commerce personalization designed specifically for resource-constrained environments. The research addresses significant implementation barriers faced by small and medium-sized e-commerce businesses through an edge-based architecture that reduces computational requirements while maintaining recommendation quality. The framework integrates federated learning techniques for distributed data processing without centralizing sensitive customer information, enabling privacy preservation while accommodating limited infrastructure capabilities. Implementation results demonstrate 26% conversion rate improvements with 41% infrastructure cost reduction compared to traditional cloud-based alternatives. The architecture leverages AWS Step Functions and API Gateway for scalable pipeline orchestration, achieving sub-50ms response times during peak traffic periods. Performance evaluation reveals 78.4% latency improvement with only 8.2% precision reduction compared to cloud-based systems. Case studies across specialty retail and home goods marketplaces validate practical applicability in commercial environments, highlighting emergent cross-categorical recommendation capabilities without explicit programming. The research establishes a comprehensive approach to democratizing advanced personalization technology, enabling businesses with limited resources to deploy sophisticated recommendation systems while maintaining operational efficiency, security compliance, and customer privacy protection.