This article has been peer-reviewed and accepted for publication in a future issue of the AAPG Bulletin. This abstract and associated PDF document are based on the authors' accepted "as is" manuscript.

Editorial Policy for Ahead of Print

Display Citation

# AAPG Bulletin

AAPG Bulletin, Preliminary version published online Ahead of Print 10 January 2022.

DOI:10.1306/12162120181

## Shale brittleness prediction using machine learning – A Middle East basin case study

### Ayyaz Mustafa1 , Zeeshan Tariq2 , Abdulazeez Abdulraheem2 , Mohamed Mahmoud2 , Shams Kalam2 , and Rizwan Ahmed Khan2

1 Center for Integrative Petroleum Research (CIPR), College of Petroleum Engineering and Geosciences, King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia
2 Petroleum Engineering Department, College of Petroleum Engineering and Geosciences, King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia

Brittleness index (BI) of rocks can help target the most suitable formation for the hydraulic fracturing stimulation in the shale reservoirs. The two most widely used approaches in the petroleum industry are based on mineralogical composition and elastic parameters for the BI estimation. However, these approaches may not be applied for all wells for BI determination due to the scarcity of mineralogical-composition and shear wave slowness data. This paper presents a machine learning (ML) approach to predict the BI using readily available well logs. Well log data were collected from three different wells that encompass a total of 2000 ft thick interval of potential shale gas formation in one of the middle eastern basins in the southeastern part of Saudi Arabia. Mineralogical composition of shale formations revealed that the shale intervals are composed of alternating high brittle and low brittle layers/zones and mainly composed of quartz, clay, feldspar, and mica. Feed-forward artificial neural network (FFANN) and adaptive neuro-fuzzy inference system (ANFIS) were employed to develop the predictive model for the BI using conventional well logs. The proposed model was tested and validated to check the consistency of the model. A total of 2007 data points were utilized in this study. ANN was found to be better than ANFIS by giving high accuracy. The proposed model was then compared with widely used models in the industry such as Jarvie et al., (2007) and Rybacki et al., (2016) on a blind dataset. Results showed that the proposed model outperformed previous models by giving less error.

## Pay-Per-View Purchase Options

The article is available through a document delivery service. Explain these Purchase Options.

 Protected Document: $10 Internal PDF Document:$14 Open PDF Document: \$24