LEVERAGING AI FOR PREDICTIVE ANALYTICS IN DATA SECURITY: IDENTIFYING AND PREVENTING ETL PIPELINE VULNERABILITIES

In the current data-driven ecosystems, ETL pipelines lend support to data Integration and Analyses used within cloud and enterprise infrastructures.Nonetheless, with time these pipelines become increasingly vulnerable to a multitude of security threats, for example, insider attacks, along with API misuse and stealthy data exfiltration.These traditional security mechanisms, typically rule-based and reactive, do not support the sophisticated, evolving threats that are woven into the very fabric of multi-stage ETL processes.The research introduces AXGNet-SHIELD (Autoencoder- XGBoost-Graph Neural Network with SHAP-based Interpretable Early-Level Detection) as a new security framework that uses AI to monitor ETL workflows for forecasting anomalies and their classifications while generating explanatory insights. AXGNet-SHIELD utilizes parallel features of unsupervised learning via autoencodersfor detecting high-dimensional irregularities in activity logs together for temporal job sequence anomaly detection followed by XGBoost event classification and GNNs for contextual risk analysis across user components and APIs.The interpretability tool DeepSHAP functions as an integrated component that displays how detection resultsShiva Kumar Vuppala https://iaeme.com/Home/