The classification of underground formation lithology is crucial for petroleum exploration and engineering as it is the basis of geological research studies and reservoir parameter calculations. Hence, there have recently been increased efforts to automate lithology classification. This is due to the rising prowess of cheap computational devices and availability of open source machine learning software libraries. This has opened avenues for the efficient analysis of large volumes of well log data with much higher accuracy. In this regard, efforts were made recently to evaluate machine learning methods to classify formation lithology by using data from Daniudui gas field (DGF) and Hangjinqi gas field (HGF). Although the machine learning algorithms utilized in the studies performed well, there is still scope for improvement in the predictive ability and scalability. The results obtained from the boosted decision tree learners, in these studies, were encouraging. Hence, we tapped into the state of the art of the boosting approach to machine learning and implemented algorithms that are scalable to large datasets. Specifically, we applied, XGBoost, LightGBM and CatBoost, which belong to the family of gradient boosted decision trees (GBDTs). We compared their performance, after combining well log data obtained from DGF and HGF, with other tree-based machine algorithms, namely, decision trees (DTs), random forests (RFs), extremely randomized trees (ERTs), AdaBoost and gradient boosting machines (GBMs). We tuned the hyperparameters and then evaluated the generated models using metrics such as the micro average, macro average and weighted average of precision (Pr), recall (Re) and F1-score (F1) on the test set. In our analysis, amongst the applied algorithms, we found that LightGBM possessed the highest metrics. Our work identifies LightGBM and CatBoost as good first-choice algorithms for the supervised classification of lithology when utilizing well log data.
