A Novel Hybrid Intelligence Framework for Predicting Strength in Sustainable Concrete Incorporating Animal Bone Ash

Abstract

The accurate prediction of compressive strength in sustainable concrete composites remains a formidable challenge owing to the inherently complex, non-linear, and multi-parametric nature of hydration reactions and pozzolanic interactions. This study presents a novel hybrid intelligence framework for multi-mixture modeling of compressive strength in lightweight concrete incorporating animal bone ash (ABA) as a partial cement replacement. A comprehensive experimental dataset comprising 45 lightweight concrete cube specimens (5 ABA replacement levels × 3 curing ages × 3 replicates) was developed, with cement partially replaced at 0%, 5%, 10%, 15%, and 20% and coarse aggregate fully replaced with pumice. Compressive strength measurements were obtained at 7, 14, and 28 days of curing. Five distinct data-driven algorithms were employed: Multilayer Perceptron (MLP), Least Square Support Vector Machine (LSSVM), Adaptive Neuro-Fuzzy Inference System (ANFIS), Hammerstein-Wiener (HW), and Autoregressive Integrated Moving Average (ARIMA). Subsequently, hybrid ARIMA-MLP, ARIMA-LSSVM, ARIMA-ANFIS, and ARIMA-HW models were developed to capture both trend and non-linear patterns inherent in the strength development process. Model performance was evaluated using the determination coefficient (R²), root mean square error (RMSE), mean absolute error (MAE), and performance index (PI), complemented by Taylor diagram visualization. Quantitative analysis revealed that the HW model demonstrated superior predictive capability for 5% replacement mixtures, achieving R² values of 0.94, 0.92, and 0.91 at 7, 14, and 28 days, respectively. The ANFIS model exhibited optimal performance for 15% replacement scenarios, with R² values of 0.93, 0.91, and 0.90 across the curing periods. Hybrid ARIMA-ANFIS and ARIMA-HW models substantially outperformed their standalone counterparts, with ARIMA-ANFIS achieving the highest overall predictive accuracy (R² = 0.98, RMSE = 0.02 N/mm², MAE = 0.01 N/mm²) for the 15% replacement mixture at 28 days. The proposed hybrid intelligence framework demonstrates significant potential as a reliable decision-support tool for optimizing sustainable concrete mixtures incorporating agro-industrial waste materials.