Machine Learning-Driven Data Quality Monitoring for Fault-Tolerant Data Pipelines

Data pipelines often face problems such as missing values, delayed updates, corrupt records, and inconsistent formats, which can lower the trust in data-driven systems. Traditional methods for data quality monitoring focus on fixed rules or manual checks, which are not flexible or scalable when data becomes large and complex. This paper introduces a novel hybrid model that combines Graph Neural Networks (GNNs), Long Short-Term Memory networks (LSTMs), and Reinforcement Learning (RL) to detect, predict, and repair data quality issues in real-time. GNNs are used to understand structural patterns between pipeline components and dataset dependencies. LSTMs handle temporal relationships to find unusual trends over time. RL helps the system choose the best correction strategy by learning from earlier feedback. The proposed model actively monitors and improves the quality of data flowing through pipelines, reducing fault risks while maintaining throughput. Simulations using open-source datasets from financial, healthcare, and sensor networks show that the model improves anomaly detection rates, reduces false alerts, and repairs data with minimal loss. The model adapts to changing environments and works with both batch and streaming pipelines. This work provides a strong foundation for making automated pipelines more fault-tolerant, self-correcting, and intelligent.