Machine learning approaches for predicting shielding effectiveness of carbon fiber-reinforced mortars

The exponential growth of communication technologies and the ubiquitous use of electronic devices raise concerns about unintended electromagnetic interference (EMI). Existing research exploring mortars incorporated with carbon fibers for EMI shielding primarily relies on empirical experiments investigating the interplay between shielding effectiveness (SE) and mixed design parameters. This research aims to establish predictive models for SE, focusing on the frequency of radiation and mortar constituents (water-to-cement ratio (W/C), fiber content, sand-to-cement ratio (S/C), and fiber aspect ratio). Employing a diverse array of machine learning algorithms, including stochastic gradient descent (SGD), gradient boosting, random forest, gene expression programming (GEP), AdaBoost, decision tree, K-nearest neighbors, and stepwise linear regression, this study aims to forecast the shielding efficacy of the mortar. Results demonstrate robust model performance, with R2 values and correlation coefficients surpassing 0.85 and 0.95, respectively, across training, testing, and validation datasets. Modeling errors, such as MSE, RMSE, MAE, and MAPE, remain within acceptable bounds for all models. The accuracy of predictions is evidenced by experimental-to-predicted SE ratios falling within the 0.5-1.5 range. Based on available data, parametric investigations conducted using the GEP-derived equation reveal a positive correlation between SE and S/C ratio, aspect ratio, fiber content, and radiation frequency, while indicating an inverse relationship with the W/C ratio. Moreover, employing Shapley additive explanation (SHAP)—a technique grounded in cooperative game theory—enhances the interpretability of SE. In summary, this study provides valuable insights into SE prediction for carbon fiber reinforced cement mortar, underscoring the efficacy of machine learning models in enhancing SE.