Unraveling the role of Raman modes in evaluating the degree of reduction in graphene oxide via explainable artificial intelligence

This paper evaluated the degree of reduction in graphene oxide, leveraging deep learning and machine learning models on over 15,000 Raman scattering spectra along with validation using density functional theory calculations. We addressed the limitations of previous studies, such as the consideration of an insufficient number of spectra as well as the lack of a comprehensive analysis of the contribution of individual Raman modes, by introducing machine learning and deep learning. Moreover, our models succeeded in predicting the carbon-to-oxygen ratio and classifying the reduction temperatures using the Raman scattering spectra as input. Employing the partial dependence plot and the feature importance, we interpreted the models and obtained consistent results on the significance of D* mode in graphene oxide. The intensity of the D* mode stands out by not only displaying the highest feature importance value for the reduction temperatures but also by correlating proportionally with the widest range of carbon-to-oxygen ratios among the various Raman modes in graphene oxide. Finally, we validated our findings through quantum mechanical calculations and confirmed the significance of the D* mode. Our study presents a comprehensive insight into the role of Raman modes in the degree of reduction as well as a precise methodology for evaluating the carbon-to-oxygen ratio of graphene oxide, a step towards its further industrial applications. [Display omitted] •Assessing graphene oxide degree of reduction by deep learning on over 15,000 spectra.•Predicting C/O ratios and reduction temperatures using deep learning models.•Unveiling the significance of the D* mode in degree of reduction through XAI.•Calculating density functional theory to validate correlation of D* mode.