پیوندهای مفید
آمار
| تعداد نشریات | 21 |
| تعداد شمارهها | 368 |
| تعداد مقالات | 3,827 |
| تعداد مشاهده مقاله | 5,118,703 |
| تعداد دریافت فایل اصل مقاله | 3,441,329 |
Comparative Evaluation of Dimensionality Reduction Methods for Lithological Unit Discrimination Based on Multi-Element Geochemical Data | ||
| Journal of Geomine | ||
| دوره 3، شماره 3، آذر 2025، صفحه 188-200 اصل مقاله (1.82 M) | ||
| نوع مقاله: Original Article | ||
| شناسه دیجیتال (DOI): 10.22077/jgm.2026.10554.1065 | ||
| نویسندگان | ||
| Soheil Zaremotlagh* 1؛ Asieh Ghanbarpour2 | ||
| 1Department of Mining Engineering, Faculty of Engineering, University of Sistan and Baluchestan, Zahedan, Iran | ||
| 2Department of Computer Engineering, Faculty of Electrical and Computer Engineering, University of Sistan and Baluchestan, Zahedan, Iran | ||
| چکیده | ||
| Identifying and discriminating lithological units from multi-element geochemical data remain fundamental challenges in petrology and mineral exploration. Geochemical datasets are typically high-dimensional, strongly correlated, and exhibit nonlinear relationships arising from petrogenetic processes such as fractional crystallization, magma mixing, and hydrothermal alteration. These complexities necessitate advanced analytical techniques for effective data interpretation. This study presents a systematic comparison of three dimensionality reduction (DR) approaches- Principal Component Analysis (PCA), Uniform Manifold Approximation and Projection (UMAP), and autoencoders (AE)- to improve the discrimination of igneous rock units. The dataset comprises 517 samples representing nine igneous rock types (granite, granodiorite, diorite, quartz diorite, gabbro, dacite, andesite, basalt, and basaltic andesite) analyzed for 22 geochemical components (10 major oxides and 12 trace elements). Following centered log-ratio (CLR) transformation and subsequent MinMax or Z-score normalization, the data were processed using the three dimensionality reduction methods and clustered with the K-means algorithm. Performance was evaluated using multiple internal and external validation indices. The combination of Z-score normalization with UMAP in four-dimensional space yielded superior clustering performance across most metrics, producing compact and well-separated clusters. In contrast, linear PCA and autoencoder methods were less effective in capturing the intrinsic structure of the data. Geochemical validation using Ce/La ratios and Harker diagrams confirmed that clusters generated by the Zscore-UMAP-Kmeans workflow correspond to coherent, petrogenetically meaningful groups. The method effectively discriminates mafic and felsic end-members, while intermediate compositions display greater variability reflecting magmatic differentiation processes. Sensitivity analysis over 30 independent runs demonstrated the stability and reproducibility of the UMAP-based approach. These findings highlight the robustness and interpretability of UMAP for revealing geochemical patterns in complex datasets, with direct implications for mineral exploration targeting and lithological mapping. | ||
| کلیدواژهها | ||
| Geochemical discrimination؛ Igneous rocks؛ Machine learning؛ Mineral exploration؛ Nonlinear dimensionality reduction؛ Unsupervised clustering؛ UMAP | ||
|
آمار تعداد مشاهده مقاله: 65 تعداد دریافت فایل اصل مقاله: 49 |
||
نماد الکترونیک
سامانه مدیریت نشریات علمی. قدرت گرفته از سیناوب