Detection of Borehole Breakout Depth in Image Logs Using Machine Learning Algorithms

This study developed a classification model for the detection of borehole breakout depths in image logs using machine learning algorithms. Approximately 1 km of image log data acquired in Gyeongju was preprocessed to make it suitable for applying machine learning algorithms. This included removing missing values and data within casing intervals and aligning the resolution among image log data — that is, vertical and horizontal resolutions were set at 0.01 m and 2.5°, respectively. Decision tree-based algorithms, random forest, and XGBoost were then applied to the preprocessed 99,090 data entries, which was divided into training (80%) and testing (20%) groups. The trained models were evaluated based on the false negative (FN) of the confusion matrix. The XGBoost model showed a superior predictive performance with 87 FNs when compared to the random forest model’s 162 FNs. The developed model can enable experts to perform detailed reviews of detected depths via the model.