Applying an ensemble support vector machine (SVM) to lithofacies prediction in the Lower Huron Shale

Source abstract

Shale lithofacies, defined by the amount of total organic carbon (TOC) and minerals composition, are valuable in identifying reservoir “sweet spots.” A study of the Lower Huron shale in the Appalachian Basin will apply an ensemble support vector machine (SVM) to classify shale lithofacies from existing well data. The method breaks down the multiple classification problem into several binary sub-classification tasks. Existing data include a limited amount of shale lithofacies interpreted directly from core tests and advanced well logging. Data from conventional log suites will be trained in each sub-classification to predict the lithofacies. The prediction results will be combined and validated in the ensemble process using the bootstrap aggregation (bagging) method to improve model accuracy.