Mohsin, Taisir and Muhsin, Duaa and Mohammed, Waffaa (2026) An Interpretable Optimized Artificial Neural Network Framework for Concrete Compressive Strength Prediction. CENTRAL ASIAN JOURNAL OF MATHEMATICAL THEORY AND COMPUTER SCIENCES, 7 (3). pp. 85-98. ISSN 2660-5309

Text CAJMTCS_938_An+Interpretable+Optimized.pdf - Published Version Download (625kB)

Abstract

Predicting concrete compressive strength is paramount to the structural safety, quality control and sustainable mix design processes. Although artificial neural networks (ANNs) are powerful nonlinear modeling capabilities , their performance is sensitive to data quality, and hyper-parameter configuration. This paper proposes a strong and full framework for automated machine learning that combines rigorous data preprocessing with full Bayesian hyper-parameter optimization. The proposed framework employs interquartile range (IQR)-based outlier removal on all eight input features (cement, blast furnace slag, fly ash, water, superplasticizer, coarse aggregate, fine aggregate, and curing age) and the target compressive strength, before removing duplicate mixture designs. After cleaning, the dataset is normalized to [−1, 1] and split into training, validation and test sets. A feedforward neural network is then searched over a wide hyper-parameter space, including network depth, neurons per layer, learning rate, training algorithm and activation function, with validation mean squared error as the objective .The optimized model achieves strong predictive performance on an independent test set : R=0.939 ,RMSE=5.304 MPa ,MAE=4.167 MPa and MAPE=14.832. Critically, the optimization process selected logsig as the best activation function, which is appropriate due to the positive, saturating nature of concrete strength growth. The study simultaneously performs statistical data cleansing and tuning of the model hyper-parameters to provide a clear and robust AI-based system for predicting the concrete compressive strength in smart construction projects.

Item Type:	Article
Subjects:	A General Works > AI Indexes (General)
Depositing User:	admin eprints
Date Deposited:	25 Jun 2026 14:07
Last Modified:	25 Jun 2026 14:07
URI:	http://eprints.umsida.ac.id/id/eprint/16615

Actions (login required)

View Item