Corrosion is responsible for up to one-third of reported pipeline incidents. Given the extensive length and limited accessibility of pipelines, operators rely on in-line inspections (ILIs) to detect and quantify corrosion defects. Although modern ILI tools provide detailed geometric descriptions of corrosion behavior, translating successive inspections into reliable growth models remains challenging. Measurement uncertainty, defect coalescence, and the joint evolution of length, width, and depth complicate learning consistent growth patterns. This study proposes a unified framework for automated corrosion growth modeling that integrates feature matching, feature engineering, and data-driven prediction into a single workflow. The framework targets defect-level modeling by jointly learning multi-dimensional corrosion growth of length, width, and depth, and by reducing correspondence-induced artifacts arising from defect coalescence and measurement uncertainty. The effectiveness of the framework was evaluated using three consecutive ultrasonic testing inspections of an API 5L X52 offshore water-injection pipeline segment. Extreme Gradient Boosting (XGBoost), a feed-forward neural network, Random Forest, and linear regression were used to assess the stability of the proposed geometric representation across learning paradigms. Integrity-oriented evaluation using Estimated Repair Factor (ERF) indicates that integrity-relevant trends are preserved. Furthermore, geometry-oriented evaluation using agreement and error analyses indicates strong performance for axial length and circumferential width, with modestly reduced performance for depth. Overall, the results suggest that the proposed framework provides a practical basis for defect-level automated corrosion growth modeling.