Machine learning–guided prediction of graphene layer numbers from electrochemical synthesis parameters

Abstract

Precise control of number of layers in graphene is critical in order to ensure consistent properties in graphene-based applications; however, traditional characterization methods are time consuming and resource intensive. In this work, we present a machine learning–guided framework in order to predict number of layers in graphene directly from electrochemical based synthesis parameters, eliminating the need for post-synthesis spectroscopic techniques. A two-stage classification method is employed, which distinguish monolayer graphene from non-monolayer samples, followed by classification of few-layer (2–3 layers) and multilayer (≥4 layers) graphene. Using a curated dataset of 441 samples extracted from the literature, and class imbalance was addressed using the Synthetic Minority Over-sampling Technique (SMOTE). Eight different machine learning algorithms were analyzed, with XGBoost emerging as the top performer in both stages, achieving an AUC of 0.98 and an overall accuracy of 95.91% in Stage 1, and an AUC of 0.81 in Stage 2. SHAP-based interpretability analysis identified centrifugation time, electrolyte molarity, and sonication time as the most influential synthesis parameters governing layer formation outcomes. The proposed approach provides a fast, scalable, and cost-effective route for synthesis-level quality assurance of graphene, highlighting the potential of artificial intelligence in carbon materials manufacturing.