Infrared (IR) spectroscopy is one of the most important methods to characterize molecular structure, and it is also an important part of the instrumental analysis experiment course for undergraduate students. In current teaching of IR spectroscopy, descriptions of how functional groups affect characteristic peak shifts are mostly qualitative, and the specific influences of different functional groups on characteristic peak shifts are rarely quantified. Machine learning (ML), an important subset of artificial intelligence (AI), enables systematic extraction of actionable insights from complex data sets. Using ML techniques to quantify the shift behaviors of characteristic peaks is a beneficial supplement to theoretical and practical teaching of IR spectroscopy. This educational study combines ML methods with IR spectroscopy and uses this integration as the teaching content of the instrumental analysis course for third-year undergraduate students majoring in chemistry. The study focuses on the characteristic peak of carbonyl (C═O) stretching vibrations. Based on IR spectroscopy teaching experiments, it guides students to construct a spectral database, develop ML models, experimentally validate the model’s prediction results, and perform SHAP (SHapley Additive exPlanations) interpretability analysis. Through this process, students successfully quantified the influences of different functional groups on carbonyl characteristic peak shifts in IR spectra. This ML-integrated teaching method can not only deepen students’ understanding of spectra-structure correlations, but also cultivate their ability to use ML techniques to solve spectrochemical problems. This work provides a foundational and replicable pedagogical framework. We encourage educators to implement or modify the provided workflow and data with proper citation and attribution, and to extend the approach to other functional groups.
Zhou et al. (Mon,) studied this question.