Predicting split-tensile strength of fiber reinforced concrete with lathe-waste steel scrap using artificial neural network

This study investigates the use of an Artificial Neural Network (ANN) model for predicting the split-tensile strength of Fiber Reinforced Concrete (FRC) incorporating lathe-waste steel fibers (LSF). A total of 288 concrete specimens were cast using a 1:2:4 mix ratio with fiber contents of 0, 1, 2, and 3% by volume. Three specimens were tested per mix and this resulted in 48 experimental datasets. The compressive and split-tensile strengths were tested at 7, 14, 21, and 28 days. Results showed that 1% fiber content achieved the highest compressive strength, while 2% yielded the maximum split-tensile strength, demonstrating the positive impact of fiber reinforcement on the tensile performance of FRC. The ANN model was developed using curing age, fiber content and compressive strength as input variables. The model was trained via the Levenberg–Marquardt algorithm with seven hidden neurons, achieved a high prediction accuracy (R² = 0.9543, MSE = 0.0234, RMSE = 0.210 and MAE = 0.1560). A performance comparison with other machine learning algorithms such as Support Vector Machine (SVM), Random Forest (RF), Gaussian Process Regression (GPR), Fine Tree and Linear Regression confirmed the ANN’s better performance in terms of accuracy. Correlation analysis and permutation-based feature importance further improved the model’s transparency and validated the input selection. The ANN model enables interpolation across intermediate mix designs (e.g., 1.25%, 1.5%, 1.75% fiber), providing a time and cost-efficient alternative to extensive laboratory testing. This study supports the application of FRC in agricultural storage structures and pavements by optimizing mechanical performance.