Machine-Learning Approach to Identify Organic Functional Groups from FT-IR and NMR Spectral Data

Abstract

Interpreting spectral data to analyze the structure and properties of unknown chemicals requires a lot of time and effort. Herein, we developed a machine-learning model that simultaneously trains on multiple spectroscopic data to identify functional groups of compounds more accurately and quickly. An artificial neural network model trained on Fourier-transform infrared, proton nuclear magnetic resonance, and 13C nuclear magnetic resonance together identified 17 functional groups with a macro-average F1 score of 0.93, outperforming the model using a single type of spectroscopy. The results indicated that training a machine-learning model with multiple spectral data can provide more accurate structural analysis when analyzing the structure of unknown chemicals, as can using multiple spectroscopy methods simultaneously.