<p>Accurate estimation of concrete strength in new structures is critical for quality control. Coring methods provide reliable results but are impractical for large-scale applications. In practice, it is impossible to test the strength of all the concrete to be cast in a structure. Testing the strength of selected samples, as commonly practiced in many countries, is insufficient for making reliable decisions about in-situ concrete quality. Non-destructive testing (NDT) methods offer effective alternatives. In this study, the strengths of selected samples are correlated with NDT measurements on corresponding structural elements, from which the potential strength is indirectly estimated at any location within the structure. The rebound hammer (RH) and ultrasonic pulse velocity (UPV) are widely used but require well-calibrated models. Conventional calibration approaches, typically based on linear regression, often fail to capture the complex, nonlinear, and multivariable nature of real-world data. This study explores the use of machine learning (ML), specifically Random Forest Regression, to enhance predictive accuracy. A dataset of 448 in-situ triplets (cube compressive strength, RH, and UPV) collected from three building sites during construction is analyzed. Complementary parameters such as concrete age, curing conditions, and element type, often neglected in traditional models, are incorporated to better capture nonlinear relationships. An extensive evaluation using statistical performance indicators, including root mean square error, prediction risk, prediction error variability, and the coefficient of determination, combined with Monte Carlo simulations, enables a probabilistic performance assessment, demonstrating that the ML-based model outperforms traditional regression methods in both accuracy and reliability. These findings highlight the potential of artificial intelligence to enhance the reliability of NDT-based diagnostics and prognostics for concrete strength estimation.</p>