LASERGAME: Leveraging Advanced Spectroscopy and Ensemble Regression for Geochemical Analysis and Model Evaluation
Authors
Hostrup, Ivik Lau Dalgas ; Houmann, Christian Bager Bach ; Østergaard, Patrick Frostholm
Term
4. term
Education
Publication year
2024
Submitted on
2024-06-13
Pages
81
Abstract
This thesis improves how LIBS data are used to predict the amounts of major oxides in geological samples. LIBS (Laser-Induced Breakdown Spectroscopy) is a fast, laser-based method that reads a sample’s chemical fingerprint. The study applies machine learning and ensemble regression models to address common obstacles: many input variables (high dimensionality), strong correlations among them (multicollinearity), and limited datasets. Two key advances are stacked generalization (stacking), which combines the strengths of multiple models for better performance, and an automated framework for hyperparameter optimization that enables systematic tuning and fair comparison. The work also compiles a catalog of models and preprocessing techniques for LIBS workflows and incorporates the results into USGS' PyHAT toolkit to extend its scientific capabilities. Together, these contributions provide a robust foundation for future geochemical analysis and planetary exploration based on LIBS data.
Dette speciale forbedrer, hvordan LIBS-data bruges til at forudsige indholdet af hovedoxider i geologiske prøver. LIBS (Laser-Induced Breakdown Spectroscopy) er en hurtig, laserbaseret metode, der aflæser prøvers kemiske fingeraftryk. Arbejdet anvender maskinlæring og ensemble-regressionsmodeller til at håndtere typiske udfordringer: mange inputvariabler (høj dimensionalitet), stærke indbyrdes sammenhænge (multikollinearitet) og små datasæt. To centrale bidrag er stablet generalisering (stacking), som kombinerer flere modellers styrker for bedre præcision, og en automatiseret ramme til hyperparameteroptimering, der gør det muligt at indstille og sammenligne modeller systematisk. Specialet samler også et katalog over modeller og forbehandlingsmetoder til LIBS-arbejdsgange og indarbejder resultaterne i USGS' PyHAT-værktøj, så dets videnskabelige anvendelighed udvides. Samlet set lægger arbejdet et robust fundament for fremtidig geokemisk analyse og planetarisk udforskning baseret på LIBS-data.
[This apstract has been rewritten with the help of AI based on the project's original abstract]
Keywords
libs ; machine learning ; chemometrics ; mars ; nasa ; usgs