Classifying Exotic Hardwood Species with Stacked Machine Learning Models and LIBS

At last week’s SciX Conference, which took place in Covington, Kentucky, Karl Booksh, a professor at the University of Delaware, delivered a talk titled, “Stacked Models and Conformal Prediction to Classify Exotic Hardwoods by Laser Induced Breakdown Spectroscopy” (1). He presented on the use of laser-induced breakdown spectroscopy (LIBS) combined with advanced machine learning (ML) techniques to differentiate among 18 classes of exotic hardwood timber. Spectra were collected from 700 samples, and while flat and one-vs-all models achieved a moderate kappa value of 0.56, performance significantly improved to 0.87 when using stacked models, which aggregate outputs from multiple classifiers into a meta-learner (1).

To quantify prediction reliability, Booksh explained how conformal prediction was applied, providing statistically valid confidence intervals that guarantee inclusion of the true class within a defined confidence level. The analysis also introduced prediction set size as a metric for evaluating overall model performance (1). Furthermore, conformal prediction identified specific timber classes and samples where the models exhibited lower accuracy, offering insights into model uncertainty and guiding future improvements in classification robustness (1).

Booksh’s research group focuses on developing in-situ chemical sensors for environmental, biomedical, and industrial monitoring. Their work integrates fiber optic surface plasmon resonance (SPR), Raman, and fluorescence sensing technologies with advanced chemometric data analysis to enhance sensitivity, selectivity, and robustness (2). His research emphasizes combining instrumental design with statistical and mathematical modeling to overcome limitations of traditional physical approaches (2). SPR sensors are being optimized for detecting proteins relevant to disease diagnosis, wound healing, and proteomic screening, as well as for identifying small molecules using molecularly imprinted polymers for applications in homeland defense and air quality monitoring (2). Fiber optic Raman and fluorescence sensors target environmental process monitoring, with long-term goals of analyzing deep-sea hydrothermal vent chemistry.

In this segment of our video interview with Booksh, he explains a major improvement in classification accuracy using stacked models for differentiating exotic hardwood species.

This interview clip is the first part of our interview with Booksh. To stay up to date with the latest coverage of the 2025 SciX Conference, click here.

References

1. SciX Conference, Stacked Models and Conformal Prediction to Classify Exotic Hardwoods by Laser Induced Breakdown Spectroscopy. SciX Conference. Available at: https://scixconference.org/2025-final-program (accessed 2025-10-13).
2. University of Delaware, Karl S. Booksh. UDEL.edu. Available at: https://www.udel.edu/academics/colleges/cas/units/departments/chem-biochem/our-people/karl-booksh/ (accessed 2025-10-13).