Technical Report: A Machine Learning Approach for Predictive Risk Assessment of Satellite Conjunction Events

This technical report presents an end-to-end machine learning pipeline for the predictive risk assessment of satellite conjunction events using publicly available Two-Line Element (TLE) data. The study addresses the problem of alert fatigue in space situational awareness by framing conjunction screening as a supervised classification task. A labeled dataset was generated via high-fidelity SGP4 orbit propagation and feature engineering of physically meaningful orbital parameters. An XGBoost classifier was trained to distinguish high-risk close-approach events from benign encounters under significant class imbalance. While the model performance highlights the inherent difficulty of the task, the analysis identifies orbital inclination differences as a key predictive feature. This work is intended as an independent technical report and research artifact to support further academic study.