The branch of artificial intelligence known as machine learning has found a growing number of applications in chemistry, from material discovery and property prediction to the calculation of force fields and potential energy surfaces. Despite the growing importance of machine learning in the chemical sciences, very few activities using these approaches have been reported in the pedagogical literature. Here we describe a pedagogical activity designed to introduce undergraduate physical chemistry students to machine learning as applied to the problem of functional group identification in vibrational spectroscopy. In this activity, students implement common machine learning binary classification algorithms to predict whether or not molecules contain a carbonyl group based on their infrared absorption spectra. As an optional extension of this exercise, students perform a multiclass classification to predict whether the carbonyl is a ketone, a carboxylic acid, or another functional group. In addition to introducing students to the basic steps in a machine learning classification analysis and to several common classification algorithms, this activity also illustrates the importance of data preprocessing and data balancing in machine learning analyses. Although designed as a standalone exercise, this activity can readily be integrated into course modules on experimental vibrational spectroscopy or computational chemistry. Reference implementations in Python and Mathematica are provided, and no software installation is required. Student surveys reveal that this activity was effective in introducing students to machine learning and stimulating student interest in the applications of machine learning to chemistry.

Machine learning for functional group identification in vibrational spectroscopy: A pedagogical activity

ACS Fall 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.