|Title||Real-time imaging of the pore structure changes in permafrost induced by thawing|
|Author||Parsa, M; Harris, J; Sherlock, R|
|Source||Mathematical Geosciences 2022 p. 1-19, https://doi.org/10.1007/s11004-022-10038-6|
|Alt Series||Natural Resources Canada, Contribution Series 20220510|
|Media||paper; digital; on-line|
|File format||pdf; html|
|Lat/Long WENS|| -78.8008 -78.5400 44.5411 44.3528|
|Subjects||Science and Technology; metallic minerals; mineralogy; Nature and Environment; gold; Prospecting|
|Released||2022 11 27|
|Abstract||Despite the ever-increasing application of machine learning (ML) algorithms in mineral prospectivity modeling (MPM), poor generalization (over-fitting) is an issue posing impediments to ML-based MPM.
This issue is partly rooted in model input variables and the paucity of mineralized zones used as labeled samples for training and validating models. This study, therefore, tries to answer the following questions to address this problem: (i) whether
using additional geologically significant labeled samples can improve MPM generalization and (ii) whether using simple binary variables instead of using multiclass and continuous variables can lessen the severity of poor generalization in MPM. A
dataset of orogenic gold mineralization in the Sturgeon Lake transect of Ontario, Canada, hosting 22 gold deposits and 46 gold occurrences, was exploited to define two suites of predictor variables describing orogenic gold mineralization. The
original suite comprised categorical and continuous variables; however, the second set was developed by converting the first suite's variables into simple binary variables. Two experiments were conducted to answer the questions raised above; while
only gold deposits were deemed labeled samples in the first experiment, the second experiment included both gold deposits and occurrences in the labeled samples. Each experiment was conducted with two sets of predictor variables, leading to four
models. Comparing the bias-variance trade-off of these models enabled the authors to draw some conclusions about MPM generalization. The results of this study can provide insights into controlling the generalization of prospectivity models.