Explainable temporal deep learning for early prediction of concrete compressive strength using IoT sensor data

Concrete compressive strength is a fundamental indicator of structural performance, yet conventional destructive testing methods require 7–28 days, limiting timely decision-making in construction projects. This study presents an interpretable deep learning framework for early prediction of concrete compressive strength using real-time multivariate Internet of Things (IoT) sensor data. The proposed approach integrates internal curing conditions (temperature and humidity), ambient environmental variables, and mix design parameters to model early-age hydration dynamics as a temporal learning problem. A Bidirectional Long Short-Term Memory (Bi-LSTM) network with an attention mechanism is developed to capture both forward and backward temporal dependencies in the curing process while identifying the most influential time intervals. To ensure interpretability, SHapley Additive exPlanations (SHAP) and attention-based analysis are incorporated to quantify feature importance and reveal critical curing stages governing strength development. The model is trained and validated using field data collected from real construction sites, enabling evaluation under realistic environmental variability. Experimental results demonstrate that the proposed framework achieves high predictive accuracy (MAE = 1.52 MPa, R 2 = 0.94), outperforming benchmark models including GRU, Temporal Convolutional Networks (TCN), and conventional machine learning approaches. Unlike existing studies that primarily rely on laboratory-controlled datasets or focus on prediction without early-stage applicability, the proposed framework enables reliable estimation of 7-day and 28-day compressive strength within the first 24–48 h of curing. This combination of real-time IoT sensing, temporal deep learning, and explainable artificial intelligence provides a robust and practical decision-support tool for smart construction systems, enabling improved quality control, optimized scheduling, and reduced reliance on destructive testing.