Pipeline Reliability Scoring with Bayesian Networks for Predictive Failure Probability in Cloud Data Architectures

Abstract

Cloud data pipelines operate across interdependent services such as ingestion layers, schedulers, compute engines, storage systems, and metadata components, which makes failure risk difficult to estimate through static health checks alone. Existing work on cloud reliability and task failure prediction highlights the value of probabilistic dependency modeling, yet practical frameworks for autonomous reliability scoring in cloud data architectures remain limited. This article presents a Bayesian-network-based reliability scoring framework that ingests operational evidence, models dependency relationships among pipeline components, estimates predictive failure probability, and converts it into a dynamic reliability score. The results show that the framework maintains strong prediction accuracy, stable reliability scoring, meaningful risk calibration, high dependency sensitivity, and responsive score updates across diverse cloud data architecture scenarios. The study shows that Bayesian reliability scoring can provide a practical foundation for proactive failure-risk assessment in cloud data pipelines.