Trade-Space Exploration With Data Preprocessing and Machine Learning for Satellite Anomalies Reliability Classification

Satellite reliability is critical to ensuring uninterrupted operations in aerospace systems, where anomalies can lead to mission failures and significant economic losses. Existing anomaly classification methods often lack scalability, interpretability, and adaptability to diverse datasets. This study introduces the Trade-Space Exploration Machine Learning (TSE-ML) framework, a comprehensive pipeline for satellite anomaly classification that optimizes preprocessing, transformation, normalization, and machine learning stages. Leveraging a Seradata dataset spanning 66 years and 4,455 satellite records, the framework systematically evaluates four data cleaning methods, four data transformation techniques, five normalization strategies, and seven machine learning algorithms across 480 configurations. The optimal configuration, comprising Iterative Imputation, FastText, Robust Scaling, and Decision Tree, achieved the highest testing accuracy of 95.74% with competitive computational efficiency. The Decision Tree model delivered superior accuracy and provided interpretability, revealing critical factors influencing satellite anomalies, such as Age Since Launch, Design Life, and Orbit Category. Stratified 5-fold cross-validation ensured robustness and generalizability of the results. The TSE-ML framework’s transparency and high performance enable actionable insights for improving satellite design, operational planning, and anomaly mitigation. Future research will focus on real-time anomaly detection, integrating satellite telemetry data, and extending the framework to other space applications. This study establishes a robust, interpretable foundation for advancing anomaly classification in aerospace engineering, addressing the dual challenges of reliability and operational efficiency.