Presentation Profile
A chemometric machine learning framework for oil classification: A California case study
Currently Scheduled: 10/14/2025 - 2:00 PM - 2:20 PM
Room: South Lobby
Main Author
Kyra Bennett - University of Houston
- Ana Vielma - University of Houston
- Joseph Curiale - University of Houston
- Thomas Malloy - University of Houston
- Jagoš Radović - University of Houston
Abstract:
Chemometrics, enhanced by machine learning, enables advanced analysis of oil geochemistry for reservoir characterization, exploration, and environmental monitoring. Using a dataset from Peters et al. (2008), we developed a decision tree model to classify California oils, seeps, tar balls, and source-rock extracts based on 19 geochemical ratios, achieving ~91% training accuracy but with signs of overfitting. Feature importance highlighted biomarkers such as bisnorhopane. Expanding to 82 new California crude oil samples with GC-MS/MS, we incorporated unconventional biomarkers, including diamondoids and carbazoles, and compared multiple ML approaches, showing promise for broader oil classification applications.
