Smart predictive models for the pull-out efficiency of fibers from cementitious matrices utilizing machine learning techniques

Fiber-reinforced concrete (FRC) is widely recognized in structural engineering for its enhanced crack resistance, durability, and sustainable performance, where the pull-out behavior of steel fibers plays a crucial role in governing fiber–matrix interaction efficiency. Despite its significance, predictive and interpretable models for fiber pull-out remain limited. This study develops two soft computing approaches, Adaptive Neuro-Fuzzy Inference System (ANFIS) and Gene Expression Programming (GEP), to predict the maximum pull-out force and slip of steel fibers embedded in cementitious composites. A dataset of 382 experimental observations, including straight, hooked-end, and spiral fibers with eight input parameters, was compiled from the literature. The models were trained, validated, and compared against an Artificial Neural Network (ANN) using statistical indices. Results show that ANFIS and ANN provided the most accurate predictions for pull-out force (R ≈ 0.98, RMSE ≈ 0.029–0.030 kN), while GEP was less effective (R ≈ 0.85, RMSE ≈ 0.078 kN). For slip, ANFIS and ANN again outperformed GEP (R ≈ 0.86–0.87, RMSE ≈ 0.12 cm vs. GEP R ≈ 0.80, RMSE ≈0.14 cm). Novel GEP-based empirical equations were derived, providing interpretable tools for engineers. Sensitivity analysis using ANFIS revealed that loading rate, fiber geometry, and aspect ratio strongly influence pull-out force, whereas fiber orientation angle and compressive strength predominantly affect slip. These findings guide optimizing FRC design and performance. Overall, the study demonstrates the effectiveness of integrating neural-fuzzy and evolutionary programming techniques for accurate, interpretable modeling of fiber pull-out behavior, supporting more efficient, durable, and sustainable structures. • Pull-out tests on steel fibers in cement matrices have been investigated. • Pull-out force and Bond-Slip predicted using ANFIS and GEP models. • Parameters affecting the performance of fiber Pull-Out were discussed. • The analytical relationship shows a strong correlation between GEP values. • The proposed model provides highly accurate estimations.