Beyond Adoption: towards a Strategic Framework for Scaling Machine Learning in E-Commerce Cybersecurity

Abstract

This study develops and validates a Machine Learning Scaling Framework (ML-SF) for e-commerce cybersecurity, addressing the critical gap between model adoption and sustainable operational scalability. Drawing on the Technology Organization Environment (TOE) framework and Dynamic Capabilities Theory (DCT), the study conceptualizes ML scalability as a socio-technical process that integrates governance, infrastructure, data readiness, continuous learning, and ethical oversight. Using three open cybersecurity datasets the CSE-CIC-IDS2018, the AI Incident Database, and the CIC-DDoS2019 the research employs rolling-window drift evaluation, Negative Binomial regression, and Interrupted Time Series analysis to examine post-adoption challenges and validate the proposed framework. Findings show that performance degradation occurs prior to detected model drift but recovers rapidly after retraining, confirming the importance of automated drift detection and retraining governance. Organizational analysis reveals that gaps in governance, data management, and MLOps maturity significantly increase incident severity, while the activation of ML-SF controls improves detection rates and reduces false positives over time. These results empirically demonstrate that scalable machine learning requires not just algorithmic innovation but institutionalized processes of monitoring, explainability, and adaptive resilience. The study contributes a theoretically grounded and empirically verified roadmap for scaling ML in cybersecurity environments. It offers actionable strategies for policymakers and enterprise leaders, including establishing AI governance boards, embedding always-on retraining pipelines, adopting federated learning for data privacy, and institutionalizing adversarial resilience as a compliance standard.