Automated Machine Learning for Lithology Prediction Derived from Seismic Data

Abstract

Abstract Lithology prediction is important for oil and gas exploration to understand the complex geological features. Machine learning techniques have shown great results in the quest to understand the complexities of seismic data. Developing an accurate machine learning model is a challenging task due to complex geological data that is limited and noisy data. The study presents Automated Machine Learning (AutoML) that can handle the intricacies of seismic data for lithology prediction. A dataset comprising 14 vertical well locations was utilized to extract seismic information along the wells and lithology data based on the gamma ray log. The extracted features serve as the input for AutoML, an automatic algorithm selection process that utilizes the Bayesian optimization algorithm to identify the best-performing model. The extracted features are used as inputs for AutoML, an automated algorithm selection process that employs Bayesian optimization to determine the best-performing model. The results indicate that the Ensemble model outperforms other algorithms, achieving an accuracy of 89%, specificity of 77%, precision of 82%, F1-score of 90%, and a Matthew Correlation Coefficient (MCC) of 79%. In conclusion, the application of a classification AutoML model has demonstrated a high level of efficacy in accurately predicting lithology derived from complex seismic data. This approach effectively captures the intricate relationships and patterns within the seismic data, enabling dependable lithology predictions.