MICA - Mineral Analysis
Authors: Mark Grujic
Working with Shiny for 1+ years
Abstract: MICA is a tool that aims to identify and group minerals that have similar chemical compositions. This is a fairly common problem; the identification of a mineral based on it's chemistry is a difficult task. Hopefully MICA can make it a bit easier!
Full Description: MICA is built from the database of minerals at webmineral.com which consists of 4722 minerals. The composition of 85 elements for each mineral is recorded in the database.
Finding similar minerals
Comparing minerals in 85-dimension space would be an challenging task. So using the UMAP algorithm (implemented in the uwot R package), we reduce the dimensionality of the data to three dimensions. Now that our data is in 3D, we can visually assess similar groups of minerals and identify naturally occuring relationships between groups of minerals.
Creating natural groups of minerals
We use the DBSCAN R package to perform density-based clustering on the reduced-dimensionality mineral data. This gives us a bunch of clusters whos constituent minerals have similar chemical compositions.
MICA shows you a list of the minerals in a selected cluster, along with their associated chemical formula.
Importance of elements within clusters
Being able to quickly assess the importance of certain elements within a group of minerals can be informative.
We use the randomForest R package to build an unsupervised random forest model, purely for investigating feature importance. The mean decrease in the Gini index for each element is ranked and displayed.
All this information comes together in an interactive Shiny app that makes Mineral Identification and Compositional Analysis easier than ever!
Category: Education
Keywords: minerals, rocks, data science
Shiny app: https://solvegeosolutions.shinyapps.io/MICA_shiny/
Repo: GitHub - Solve-Geosolutions/MICA_shiny: Mineral Identification and Compositional Analysis
RStudio Cloud: Posit Cloud
Thumbnail:
Full image:
