EURECA-PRO Invited Lecture at TUC:
Dr. Dimitrios Meimaroglou
“A Machine Learning approach in product engineering for the prediction of the properties of molecules”
4th of June, 2024 @ TUC Campus, Chania, Greece | on-site & on-line via Zoom
Invited Lecturer: Dr. Dimitrios Meimaroglou | Université de Lorraine, France
Associate Professor in the Environmental Chemistry Department (ENSIC)
Abstract
This work investigates the use of machine learning (ML) methods for the prediction of thermodynamic properties (i.e. enthalpy and entropy of formation) of molecules from their molecular descriptors.
Although quantum chemistry (QC) or group contribution (GC) methods have been commonly employed to calculate these properties, they have shown limitations in their applicability to more complex or larger chemicals and/or computational costs. Inversely, ML methods, based solely on data, have already demonstrated their ability to tackle complex problems in other fields when classical approaches fail or are inefficient. However, their implementation is often mistakenly considered as plug-and-play, overlooking or underestimating the effect of the different choices that are adopted along the development and application of these techniques on the final model performance.
Accordingly, the contribution of this work is rather a methodologically-driven investigation of these effects and an attempt to understand how to follow an optimal path – if any – during the implementation of ML techniques to similar problems.
