Boosting machine learning algorithms for porosity prediction in sandstone gas reservoirs

Porosity is one of the crucial properties of a reservoir, which relates to the volume of void space in a rock. The pivotal petrophysical attribute of a rock, with profound implications for hydrocarbon reserve estimation, reservoir evaluation, and petroleum economics, is its porosity. In the laboratory determination of porosity, the acquisition of high-quality samples is imperative. However, the preparation of core samples is encumbered by various constraints, including the challenge of difficulty, expense, and time consumption. That’s why the development of predictive models to determine the reservoir porosity of rocks seems to be an attractive research. Considering the rising interest towards the development of data-driven estimation models, this study utilized five boosting machine learning regression algorithms: Adaptative Boosting (Adaboost), Category Gradient Boosting (Catboost), Light Gradient Boosted Machine (LightGBM), Gradient Boosting Decision Tree (GBDT), and Extreme Gradient Boosting (XGBoost) to determine porosity from a sandstone gas reservoir. Hyper-parameters were tuned within an appropriate tuning search range to optimize the model performance along with Grid Search. A database containing 776 well-log readings was employed to formulate and validate the models. The assessment of model effectiveness involved the utilization of performance metrics such as the coefficient of determination $$\:\:\left({R}^{2}\right)$$ , root mean square error (RMSE), and average absolute percentage error (AAPE). Based on the accuracy of the models, a comparative analysis was conducted, leading to the conclusion that the GBDT model outperforms other developed models significantly, as evidenced by an $$\:{R}^{2}$$ value of 0.92, an impressively low RMSE of 0.0008, and an AAPE of 2.91 for the test dataset.