Scalable Machine Learning Pipeline for U.S. Visa Approval Prediction Using AWS

Abstract: The U.S. visa approval process involves complex, high-volume decisions shaped by applicant and employer attributes. This essay offers a comprehensive Machine Learning Operations pipeline for automated visa approval prediction using the Easy Visa dataset (25,480 records, 12 features). The system integrates automated data validation, schema enforcement, class imbalance handling, and feature engineering within a reproducible pipeline deployed on Amazon Web Services (AWS) . Several classifiers were assessed under the same conditions, the KNN classifier achieved 96.83% accuracy and the highest F1-score. The pipeline incorporates continuous monitoring, drift detection, and automated model promotion, transforming a static classifier into a sustainable decision-support framework. Results confirm the viability of end-to-end Machine Learning Operations integration for administrative classification tasks.

Keywords: MLOps, visa prediction, Amazon Web Services, data drift detection, binary classification, cloud deployment, K-Nearest Neighbors.