Laser-Induced Graphene (LIG) from Different Polymer Precursors Predicted Using Machine Learning

The conversion of polymers to laser-induced graphene (LIG) is a facile and low-cost process to create a patterned conductive nanomaterial for applications such as sensing and energy storage. However, the complexity of the laser conversion process means that it is difficult to predict the effect of different laser parameters, especially for different polymer precursors. This research leverages advanced machine learning models to optimize and predict the laser parameters necessary for polymer conversion into LIG. Data to train the models was collected experimentally and extracted from the academic literature, after which preprocessing techniques were applied. A major input parameter studied in this work is the type of polymer precursor. To this end, we transformed the molecular structures of different polymers from their Simplified Molecular-Input Line-Entry System (SMILES) representation into molecular fingerprints, descriptors, and tokenized representations using RDKit and a pre-trained BERT model. A variety of predictive models including fully connected neural networks, random forest, gradient boosting, and XGBoost were trained, evaluated, and optimized. The cornerstone of the analysis was a specialized testing of feature importance across three distinct SMILES representations, illuminating the complex interplay between molecular structures and polymer properties. The superior performance of the random forest model highlights the vital importance of feature selection and optimization. This creates a comprehensive understanding of influential features in predictions including both laser parameters and different molecular representations. By integrating traditional molecular descriptors and advanced machine learning techniques, this research offers a robust framework for predicting LIG properties, potentially reducing extensive experimental needs. The results have broad implications in material science and flexible electronics.