Explainable ML: Machine learning interpretability methods
applied to spectroscopic data
Sylvie Roussel presents Ondalys’ latest work on the interpretability of machine learning models built on NIR spectra within the Mediterranean Chemometrics community in Porquerolles.
How can the black box effect can be avoided when developing advanced Machine Learning methods such as Support Vector Machines, Artificial Neural Networks and Boosting methods?
Ondalys team worked on the interpretability of ML models developed on near-infrared spectroscopic data, in other words, the “post-hoc explanation” of ML models.
The objective was to develop methods to:
- understand and interpret the spectral models,
- avoid overfitting
Several regression methods (SVM, ANN, XG-Boost) were trained and compared with classical PLS calibration models. The idea was to adapt Explainable AI (X-AI) methods to spectroscopic models to increase the reliability of predictive model results.
Discover the results of this study
The Colloquium Chemiometricum Mediterraneum is a scientific event that discusses the latest innovations in Chemometrics, with presentations by French, Spanish, Portuguese, and Italian chemometricians in their respective own languages. The CCM is a place to intiate new international collaborations within the chemometrics community.
Learn more about CCM 2025 on the Colloquium website
