Prediction of Compressive Strength of Iron Tailings Concrete Based on Interactive Cross-Channel Multilayer Perceptron Neural Network

Abstract Iron tailings, a by-product of iron ore refining, pose significant environmental challenges due to massive stockpiling worldwide, especially in major iron-producing countries. Reusing iron tailings in concrete not only mitigates environmental risks but also alleviates the shortage of natural sand resources. However, the multifactor and multiproperty coupling characteristics in the mix ratio of iron tailings concrete make general prediction models difficult to fit accurately. To address this issue, this study designed the interactive multichannel embedding module (IMCEM) and cross-mixing module (CMM), and constructed the interactive cross-channel multilayer perceptron (ICM-MLP) model by combining them with the multilayer perceptron (MLP). Based on the multifactor-coupled data set of 588 iron tailings concrete samples constructed in this study, the prediction accuracy of ICM-MLP was compared and validated against the latest models in the field. The results demonstrated that ICM-MLP achieves the best prediction performance, outperforming the second-best machine learning model support vector regression (SVR) and the unoptimized MLP by 8.20% and 15.32% in the coefficient of determination ( R 2 ), respectively. Specifically, ICM-MLP attained an R 2 of 0.731, with mean absolute error (MAE), mean squared error (MSE), and mean absolute percentage error (MAPE) values of 3.552, 24.790, and 9.071%, respectively. Ablation experiments further confirmed that the incorporation of CMM and IMCEM significantly enhances the model’s ability to capture the complex multifactor and multiproperty relationships in iron tailings concrete mix proportions. This study provides a novel, effective, and reliable deep learning–based approach for high-accuracy prediction of iron tailings concrete compressive strength, offering valuable insights for sustainable construction material design.