Sarcouncil Journal of Engineering and Computer Sciences

Abstract: The burden of considerable challenges involved in deploying predictive systems across global markets is increasingly growing for the pharmaceutical industry's reliance on machine learning models for commercial analytics. Differences in healthcare systems, regulations, patient demographics, and prescribing practices pose serious challenges to the transferability of models and often lead to reduced performance and biased predictions when using out-of-area models. In this article, we offer a full framework for improving model portability using novel computer science methods: federated learning, domain adaptation, and transfer learning. By analyzing commercial datasets from multiple geographic regions, key sources of model degradation are identified, including covariate shift, feature distribution misalignment, and regulatory constraints. The proposed framework addresses these challenges through privacy-preserving collaborative training, domain-invariant feature representations, and causal inference methods that ensure robust performance across diverse markets. Meta-learning strategies enable rapid adaptation to new markets with minimal data requirements, while continuous monitoring systems detect performance degradation and trigger appropriate interventions. Case studies from multinational pharmaceutical companies demonstrate the framework's effectiveness in maintaining model accuracy while ensuring compliance with regional data protection regulations. The integration of fairness-aware algorithms and bias mitigation techniques ensures equitable model performance across different patient populations and healthcare systems. These findings provide pharmaceutical organizations with actionable strategies for developing resilient, globally deployable AI systems that support commercial decision-making while respecting regional variations and regulatory requirements.