Leveraging machine learning for predicting the photocatalytic performance of a g-C3N4/CdS/MoS2-based heterostructure nanocomposite

Abstract

A recent study showed that the g-C₃N₄ supported transition-metal di-chalcogenide (g-C₃N₄/CdS/MoS₂) nanocomposite faces limitations in enabling the development of efficient photocatalysts for visible light degradation of water contaminants due to an intensive, time-consuming experimental process. In this work, we leveraged machine learning (ML) modeling to evaluate the performance of photocatalytic degradation of methylene blue (MB) using the g-C₃N₄/CdS/MoS₂ heterostructure nanocomposite. Four different ML algorithms (RF, DT, SVM, and NN) have been used to develop a regression model for predicting the photocatalytic performance of the g-C₃N₄/CdS/MoS₂ heterostructure nanomaterial. The manually curated dataset consists of six independent features for training and testing the models. The best-trained ML models are RF and NN, displaying the highest prediction accuracy values of R² = 0.7014/0.6864, R = 0.844/0.8285, and RMSE = 4.1963/4.3002 as predictive models, suggesting 83.5% and 83.7% efficiency for photocatalytic degradation of MB under 180 minutes of sunlight irradiation. The predicted photocatalytic efficiency was validated against experimental results, demonstrating 86% degradation of MB under optimal conditions. The ML model training and testing observations align with these findings, maintaining an error margin of 5%. The trained ML model, based on a manually curated dataset of the g-C₃N₄-supported CdS/MoS₂ heterostructure, significantly reduced the resource-intensive experimental process and accurately predicted the photocatalytic efficiency of g-C₃N₄/CdS/MoS₂ with high precision.