Pipeline integrity assessments in upstream pipelines is a challenging activity that requires considerable amount of data and its proper interpretation. An important factor, usually ignored during the data collection process, is a data integration and data correlation analysis. This process becomes of high importance when trying to collect meaningful and representative pipeline information. The Success criteria of the integrity assessment highly depend on the quality of the data gathered, and the appropriate selection of significant variables in the corrosion mechanism. This approach aims to improve the decisionm making framework on the internal corrosion of pipelines. This report encompasses a statistical analysis of oil pipelines metal loss due to localized corrosion. Different Machine Learning (ML) methods and statistical approaches, like Decision Tree, Logistic Regression and Random Forests were compared to identify the statistical significance of different predictor variables, such as, geochemical parameters, operation parameters, and mechanistic simulation outputs. The results showed that meaningful optimization of significant predictor variables enhance the ML model prediction accuracy.