A novel XRF-based lithological classification in the Tarkwaian paleo placer formation using SMOTE-XGBoost
Files
(Published version)
Date
2022
Authors
Ibrahim, B.
Ahenkorah, I.
Ewusi, A.
Majeed, F.
Editors
Advisors
Journal Title
Journal ISSN
Volume Title
Type:
Journal article
Citation
Journal of Geochemical Exploration, 2022; 245(article no. 107147):1-14
Statement of Responsibility
Conference Name
Abstract
Geological processes such as weathering and metamorphism normally lead to the destruction of primary physical features on rocks, making it difficult to accurately classify lithologies during manual core logging. As a result, manual core logging remains a subjective and inconsistent process. Machine learning techniques have shown tremendous success in lithological classification, but their performance has significant drawbacks, particularly in highly imbalanced data domains and complex geological terrain. Because there is an increasing demand for accurate and timely lithological classification, this study proposes an objective hybrid approach based on synthetic minority oversampling technique and extreme gradient boosting (SMOTE-XGB) for lithological classification within the Tarkwaian paleo placer formation using assay data obtained through X-Ray Fluorescence (XRF) analysis.
Five other intelligent approaches were developed for comparative purposes to further evaluate the performance of the proposed approach, using accuracy (Acc), area under the curve (AUC), precision, sensitivity, and F1 score as evaluation metrics. The statistical results have proven that the proposed SMOTE-XGB approach is more efficient at classifying the Tarkwaian lithologies than the other machine learning models used in this study. When SMOTE was applied to the imbalanced dataset, the results revealed a significant improvement in the classification performance of the minority classes, particularly the mineralised lithologies. To that end, this research has shown that applying SMOTE to an imbalanced dataset can improve the prediction efficiency of machine learning models.
School/Discipline
Dissertation Note
Provenance
Description
Access Status
Rights
Copyright 2022 Elsevier
Access Condition Notes: Accepted manuscript is available open access