International Journal of Electrical and Computer Engineering (IJECE) Vol.
No.
October 2025, pp.
ISSN: 2088-8708.
DOI: 10.
11591/ijece.
Enhancing concrete sustainability: a neural networks hybrid optimization approach to predicting compressive strength using supplementary cementitious materials EsraAoa Alhenawi1.
Ayat Mahmoud Al-Hinawi2.
Zaher Salah3.
Omar Alidmat1.
Esraa Abu Elsoud1.
Raed Alazaidah1.
Bashar Rizik AlSayyed4 Department of Computer Science.
Faculty of Information Technology.
Zarqa University.
Zarqa.
Jordan Department of Allied Engineering Sciences.
Faculty of Engineering.
The Hashemite University.
Zarqa.
Jordan Department of Information Technology.
Faculty of Prince Al-Hussein Bin Abdullah II for Information Technology.
The Hashemite University.
Zarqa.
Jordan Department of Civil Engineering.
Faculty of Engineering.
University of Jordan.
Amman.
Jordan
Article Info
ABSTRACT
Article history:
This research evaluates the implementation of advanced machine learning methodologies for concrete mix design to achieve better predictive models and sustainable outcomes.
This study develops a hybrid optimization approach by combining dung beetle optimizer (DBOA) and firefly algorithm (FLA) to optimize hyperparameters for convolutional-recurrent neural networks in order to correctly predict concrete compressive strength when using supplementary cementitious materials (SCM.
Shapley additive explanations (SHAP) provide feature significance analysis, which ensures that the model produces understandable conclusions supported by empirical The findings demonstrate that this method enhances the predictive accuracy of strength analysis, along with offering critical insights about SCM usage in order to improve sustainable construction methods.
The model proves suitable for integration into actual concrete mix design and quality control systems because it achieves both computational speed and passes validation tests on distinct datasets.
The research creates foundations for upcoming studies about multimodal learning enrichment and deals with ethical concerns in construction site safety when using machine learning Received Aug 30, 2024 Revised Jun 19, 2025 Accepted Jun 30, 2025 Keywords:
Firefly algorithm optimization Gas emissions Greenhouse Supplementary cementitious Sustainable construction This is an open access article under the CC BY-SA license.
Corresponding Author:
Zaher Salah Department of Information Technology.
Faculty of Prince Al-Hussein Bin Abdullah II for Information Technology.
The Hashemite University Zarqa.
Jordan Email: zaher@hu.
INTRODUCTION
The building industry has implemented many cement manufacturing greenhouse gas emission reduction measures .
, .
Similar to burning gasoline, the manufacture of hydraulic cement accounts for 79% of world carbon dioxide emissions .
Ae.
To lower CO2 emissions, cement mixes can include waste or industrial flows as multi-component binders .
, .
Ae.
One way to cut greenhouse gas emissions by 47% to 5% is to use industrial waste instead of cement, such as ground granulated blast furnace slag (GGBS) from blast furnace iron ore extraction .
, .
One ton of GGBS production has a worldwide environmental emission factor of 0.
143 t CO2-e/ton, which is less than the standard established by most Journal homepage: http://ijece.
ISSN: 2088-8708
nations and international organizations.
Cement uses 9 t CO2-e/ton, although concrete mixes are made using transportation technologies .
About 60% of the 530 million tons of global slag (GGBS) is used in construction .
, .
GGBS
has been studied on various concrete and mortar types .
, .
, .
Another study .
found that MS and GGBS synergistically improved concrete resistance to unfavorable consequences.
According to a different research .
, resistance and durability without PC were best boosted by 20% MBS made from rice husk and by using slag powder instead.
By converting fly ash and powdered blast furnace slag, waste glass-derived nano powder, into alkali-activated mortars, drying shrinkage is decreased and wear, freeze-thaw cycles, and sulfuric acid resistance are enhanced .
According to other studies.
GGBS should be used in place of 20% of cement before sample workability and porosity cause resistance to drop .
, .
The workability, bleeding, heat of hydration, corrosion, porosity, and permeability of both fresh and hardened concrete are all improved by GGBS in cement .
Ae.
Due to greater particle distribution.
GGBS improved concrete workability by 40% .
Cement paste fills aggregate micro-spaces to reduce internal friction between concrete components, improving workability .
, .
Tensile strength, a vital road concrete property, has design parameter constraints that should be controlled to minimize cracks .
Concrete shrinkage depends on cement, water, aggregates, air dryness, and rising temperature .
Cement dose.
C3A content, gel component, and alkali content affect shrinkage .
Early setting reduces cement paste volume by 1% of dry cement volume .
, .
Shrinking values fall below zero.
Concretes with a lot of binder can reach 6 mm/m .
Size decreases by 5% after one month, 60% after three months, and 75% after a year .
W/C ratios cause concrete pores to expand, which increases shrinkage.
Additional aggregates prevent shrinkage because they are coarser .
More than 50% relative humidity and 800 ppm CO2 increase shrinkage rates because CO2 dissolves and eliminates hydrosilicate hydration products .
The carbonation process is accelerated by high external CO2 concentrations, lowering the pH from 12Ae13 to less than 9 and rupturing the reinforcing bar's passivation layer, which causes corrosion .
Ae.
As the largest component of concrete, aggregate usage has the greatest environmental impact.
Annual international building aggregate demand exceeds 10 billion tons .
Ae.
Road asphalt pavement (RAP) aggregates .
Ae.
, quarry sand (QS) .
, recycled aggregate concrete (RAC) from building demolitions .
, .
, and ecological mortar that replaces natural aggregates with glass waste .
have all been the subject of cadaveric research.
ACBFS is another air-cooled road concrete aggregate source .
, .
It has been demonstrated that natural aggregates made from blast furnace slag may be used in asphalt .
For concrete structures, the Japanese Guide suggests a 2060% combination of natural and fine materials .
The reference standard for blast furnace slag concrete aggregates in our nation has been SR EN 12620 since 2003 .
It is possible to incorporate steel industry waste materials into road concrete mixtures, but it is important to assess how they affect both fresh and hardened concrete .
Utilizing innovative technologies and reusing synthetic materials can help the environment by reducing manufacturing costs and conserving non-renewable resources .
A low-carbon circular economy is a global goal, and the building industry is promoting green alternatives and minimizing environmental harm caused by the cement industry.
The production of concrete and cement produces almost no emissions.
Nine metric tons of CO2 must be released by the cement industry for every metric ton of cement produced.
There are major repercussions linked to the global emissions from the building industry.
Therefore, one of the greatest approaches to reduce greenhouse gas emissions is to replace cement with similar materials .
Supplementary cementitious materials (SCM.
are cost-effective and ecologically beneficial alternatives to cement.
Most SCMs are pozzolanic, which improve concrete microstructure, late strength, and CO2 emissions .
Fly ash (FA) from coal power stations is the most common SCM in cement-based materials.
Industry standards recommend replacing cement with FA by 1030% .
, .
High volume fly ash (HVFA) concrete has the best mechanical properties and durability, making it a sustainable construction material .
Ae.
Following 56 and 91 days of mixing FA with cement and replacing 50% of the cement with FA at a W/C ratio of 0.
Bouzoubay et al.
achieved compressive strengths of 32.
2 MPa and 2 MPa.
These values were greater than ordinary portland cement (OPC) concrete's 0.
According to Mardani-Aghabbaglou and Ramyar .
, after 180 days, 60% FA replacement increased compressor strength by 1518% above OPC.
According to Chen et al.
, drying shrinkage was decreased by 2330% when FA was substituted for 5080% of the cement.
According to Myllauer et al.
, 70% FA successfully decreased alkali-aggregate reactions.
According to Wang et al.
, to reduce the porosity of concrete, 30% by weight of coarse aggregate should be used for FGPS and 15% for FA.
Additionally.
Wang et al.
discovered that adding 30% FA to panel concrete improved its permeability and compressive strength with Int J Elec & Comp Eng.
Vol.
No.
October 2025: 4965-4982
Int J Elec & Comp Eng
ISSN: 2088-8708
Granulated ground blast furnace slag (GGBFS) blast furnace iron waste is frequently used as a cement alternative.
Glassy blast furnace slag is quenched and cooled to form GGBFS, which binds well.
Crossin .
found that GGBFS as an SCM reduced greenhouse gas transparency by 47.
After 28 days of curing, the mechanical strength of ternary mixed cement containing a larger percentage of GGBFS was greater than that of OPC 22, .
, .
, .
Due to the diluting effect.
Lim et al.
discovered that ternary mixed cement containing GGBFS had a lower peak compressive strength than OPC.
However, it did better After 28 to 365 days of therapy.
OPC and acquired more between 28 and 91 days .
Ae.
Cheah et al.
investigated a ternary mixed cement mortar that contained cement.
GGBFS, and ground coal bottom ash (GCBA).
They discovered that the mechanical and physical characteristics of cement mortar were enhanced by 40% GGBFS and 5% GCBA without SP and 40% GGBFS and 10% GCBA with SP .
Research on waste concrete aggregates and ternary cementitious ingredients like silica fume and GGBFS revealed that adding GGBFS to 25% of cement increased the GW P value.
According to Weise et al.
, the 30% weight metakaolin mix consumed the most CH between days 28 and 56, which had an impact on strength.
Using SCMs in concrete improves cement reaction, but metakaolin content increases efficiency .
SCMs like FA and GGBFS can enable a low-carbon circular economy in building.
These materials reduce greenhouse gas emissions and increase concrete structure efficiency and duration, encouraging construction sustainability.
Impermeability in concrete affects resistance to water-soluble chloride ions.
CO2, and sulfate, which impair concrete durability.
These chemicals can infiltrate concrete, degrade it, and limit its lifespan, making them global threats .
Ae.
Concrete porosity, which affects its permeability, depends on its size, shape, and interfacial transition zone.
Recycled fine aggregate (RFA) %.
W/C ratio, and mineral admixtures directly affect recycled fine aggregate concrete (RFA.
According to research, admixtures and additives optimize concrete's microscopic structure, boosting its impermeability .
Ae.
Concrete porosity is highly dependent on RFA quantitative properties.
RFA has fewer physical properties than natural fine aggregates.
Because RFAc uses more RFA, its performance is usually poorerRFA particles absorb gaps and microcracks from the previous mortar because of their poor grading, which makes the concrete porous.
Thus.
RFAc replacement increases the water vapor transmission barrier .
Ae.
Because RFA is porous and absorbs water.
SCC mixtures absorb more water with a higher RFA substitution However.
RFAc's permeability is lower in a sulfate environment than in regular concrete, and its water absorption is 25% lower .
The W/C ratio is crucial to concrete preparation and permeability resistance.
W/C ratio of 0.
65 reduces impermeability compared to 0.
Because a larger W/C ratio slows RFAc cement hydration, pores form and concrete compactness decreases .
, .
, .
FA can replace silicate cement in construction due to its pozzolanic characteristics .
Concrete that contains fine fly ash has better hydration, zeolite, accumulation, and nucleation .
Concrete density rises as a result of the early hydration response of FA, which increases the quantity of hydration products in the pores .
The permeability of self-compacting concrete containing 10% FA and 100% RFA was lower than that of control mixes devoid of FA .
Due to RFA's poor physical and chemical characterization and RAC's weak interfacial transition zone (ITZ), alternative mixing strategies have been introduced.
This comprises the optimal triple mixing technique (OTM), triple mixing method (TM) .
, and double mixing method (DM) .
To conclude.
OTM staff add SPs, water-reducing agents, in varied orders.
SP is added to the mix with other gelling components to increase gelling and ITZ proportion.
This method uses the zeolite effect more effectively.
hence the RAC has an 8-permeability rating .
By choosing the correct materials and ratios and improving procedures, the building industry can waterproof and strengthen concrete.
This provides durable RFA beats NFA due to fine aggregate's poor water absorption.
RFA's water content must be evaluated before integration because higher absorption affects concrete's mechanical qualities.
AD.
OD, and SSD RFA methods were tested for RFAc permeability.
The study found that RFA permeability resistance in concrete follows the order SSD > AD > OD, while for recycled coarse aggregate, the order is reversed (SSD < AD < OD) .
Compared to RCA.
RFA restricts water more because its particles are smaller and its specific surface area is larger .
Concrete pore size is significantly impacted by RFA integration, and permeability is decreased by raising the W/C ratio and RFA replacement rate, as shown in Figure 1.
Saturated surface-dried RFA, improved mixing, and fly ash addition, on the other hand, consistently lower RFAc permeability.
Fly ash with two pozzolanic minerals, silica fume, and metakaolin, together, increases the impermeability of RFAc.
Table 1 lists the variables influencing RFAc sealing.
In conclusion, although increased water absorption is a problem with RFA, its permeability and durability may be significantly increased by controlling its moisture content, merging SCM, and using the precise mix proportions for RFAc.
This paper presents a unique hybrid optimization technique that combines the dung beetle optimizer (DBOA) and firefly algorithm (FLA) to improve the hyperparameters of a convolutional-recurrent neural Enhancing concrete sustainability: a neural networks hybrid optimization A (EsraAoa Alhenaw.
A ISSN: 2088-8708 network in addition to using traditional approaches.
Through their combination, the approach performs a strong global search using FLA alongside precise local search optimization done by DBOA.
Our method combines conceptual novelty by applying biological concepts to material scientific applications when predicting concrete compressive strength as an essential factor for sustainable construction.
In addition, this research aims to address the critical gap in sustainable construction practices by significantly reducing the CO2 emissions associated with conventional cement production.
The study examines how concrete's mechanical characteristics change after supplemental cementitious materials (SCM.
such fly ash (FA) and ground granulated blast furnace slag (GGBS) are added.
Figure 1.
RAC ycOyca is the weight percentage of SCM's residual water in the gelling material overall.
ycOyca is 60Ae80% of the product of RA's weight and water absorption.
and ycOyca is the result of mixed water OeycOycaOeycOyca Table 1.
Numerous elements' effects on RFA concrete's (RFA.
impermeability Factor RFA moisture level Mineral additives Enhanced triple mixing technique (OTM) RFA proportion Water-cement ratio Impact Positive Positive Positive Negative Negative Variation Increase Increase Increase Research results show that our hybrid algorithm optimized neural network achieves precise compressive strength prognosis and provides sustainable mixtures with lower environmental effects.
The manuscript establishes multiple new findings for both construction engineering science and environmental sustainability research.
This manuscript proves the effectiveness of combining bio-inspired optimization methods for material assessment purposes in an atypical application domain.
The research demonstrates that SCMs have the potential to reduce greenhouse gas emissions substantially.
The predictive features of this research will result in new modeling capabilities for planning and simulation tools to integrate directly into construction processes, thus enabling real-time, environmentally responsible material selection choices.
We will begin by providing an overview of the current state of cement manufacturing and its environmental impacts, highlighting the use of SCMs as a sustainable alternative.
Following this, we will delve into the methodology section, outlining the hybrid optimization techniques and neural network models employed to enhance the predictive accuracy of concrete compressive strength.
Subsequent sections will discuss the experimental setup, results, and a detailed analysis of the findings, focusing on the performance and generalizability of the model.
Finally, we will conclude with a discussion on the implications of our research, future research directions, and the potential integration of our findings into a concrete mix design and quality control systems.
PREVIOUS WORK
Numerous studies on concrete technology have identified important factors and enhanced the prediction of concrete strength (CS) through the use of cutting-edge machine learning algorithms.
For fly ash-based geopolymer concrete CS, decision tree algorithms, bagging, and AdaBoost regressors were used .
The bagging model made the best predictions, with an R-squared of 0.
The most crucial CS characteristics were identified by a comprehensive sensitivity investigation, supporting the environmental sustainability of geopolymer concrete.
Feng et al.
developed an intelligent CS prediction method based Int J Elec & Comp Eng.
Vol.
No.
October 2025: 4965-4982
Int J Elec & Comp Eng
ISSN: 2088-8708
on AdaBoost.
Using AdaBoost, they outperformed artificial neural network (ANN) and support vector machine (SVM) on more than 1,030 case sets.
In their article, they used sensitivity analysis and 10-fold cross-validation to calculate accuracy and precision.
Research on recycling aggregate concrete (RAC) .
using ANN symbolic learning and GEP.
Sensitivity study revealed CS-affecting factors, and GEP performed better than ANN.
Additionally, the study proposed that bagging and boosting might enhance prediction.
SFRC beam shear resistance was predicted by many machine-learning algorithms .
The dependent variable was best predicted by XGBoost using tried-and-true machine learning techniques.
Input parameters are anticipated in the study.
Chen et al.
GBRT predicted concrete-FRP bond resistance.
The best method and model.
GBRT, predicted the issue.
ANNs and genetic algorithms (Ga.
or particle swarm optimization (PSO) predicted CES bond strength .
Test sensitivity analysis identified crucial variables.
CS was evaluated at high temperatures using AdaBoost.
Random Forests, and decision trees .
Highly cement sensitive.
Gaussian process regression (GPR) with the Matern32 kernel function predicted high-performance concrete's CS better than ANNs .
In the sensitivity study, cement concentration and testing age were crucial.
CS was predicted by Random Forest using field and lab data .
Field-trained models showed improved accuracy, indicating that several data sources reduce Decision tree and gradient boosting tree models were used to examine the bending performance of FRP-reinforced concrete beams .
Beam depth, flexural reinforcement area, and assessment metrics supported the gradient-boosting tree model.
ANNs, decision trees.
Bagging, and gene expression programming predicted CS .
Bagging was most accurate at 0.
95 R-squared.
The XGBoost model with manually selected features performed well in .
when estimating CS based on concrete composition and cure period.
According to the study, a decrease in dimensionality helps the support vector regression model.
Machine-learning approaches were used to forecast setting time and strength development in Ordinary Portland cement binders .
The results were comparable to ASTM test Finally, .
tested ANN, boosting, and AdaBoost ensemble machine-learning approaches for geopolymer concrete CS prediction using high-calcium fly ash.
Due to its accuracy, the boosting approach was acknowledged, and these findings suggest ensemble methods for enhancing concrete for sustainable This study aims to improve concrete property forecasts.
Improving predictive models requires understanding the intricate relationship between material components, ambient environments, and concrete physical properties.
In the construction industry's quest for efficiency and sustainability, this study will enhance the prediction model's accuracy and adaptability through an analysis of environmental impacts and mechanical behaviors.
PROPOSED METHODOLOGY
This study uses extensive analysis, data management and manipulation methods, advanced artificial neural network architecture, and hybrid optimization algorithms.
For optimal dataset use and high-quality findings, the dataset was cleaned and preprocessed using concrete properties.
Data was normalized to reduce skewness in feature scales.
The neural network architecture served as the foundation for the built predictive model, which included convolutional and recurrent layers to extract the temporal relationships of the data in order to guarantee an accurate forecast of concrete strength.
Because it provided a solid basis for further enhancements, this model arrangement was initially appropriate for testing performance.
The study advises creating FLA and dung beetle optimizers.
This procedure improved model hyperparameter values more than normal.
FLA effectively explores parameter space via global search DBOA founds the local optimum.
A large solution space and reliable model prediction are guaranteed by comprehensive search approaches.
Figure 2 shows the important methodology's application and linkages.
Finally, the FLA DBOA model's algorithm flow and predictive model integration are displayed in the following Figure 2.
In our methodology, we conducted a comprehensive dataset preprocessing to ensure the integrity and consistency of our model.
The implementation of normalization standardized numerical values to match ranges while maintaining stability and convergence, through which the IQR method detected outliers to remove anomalous data that affected the analysis.
The Min-Max scaler method was implemented for feature scaling in order to equalize the effects of each input feature upon model predictions.
Shapley additive explanations (SHAP) assessed the influence of different variables on concrete compressive strength predictions by assigning value weights to each contributing feature in the prediction.
The predictive model gained both enhanced interpretability and transparency when using SHAP, which revealed vital features together with clearer explanations to increase the reliability of predictions.
Enhancing concrete sustainability: a neural networks hybrid optimization A (EsraAoa Alhenaw.
A ISSN: 2088-8708 Figure 2.
Proposed scheme The model performed its training on hardware optimized for GPU computing to decrease processing time while attempting to scale the system.
The design enables our system to process extended datasets effectively thus enabling practical applications that need immediate, precise outcomes.
The implemented model relies on state-of-the-art computer tools and methods to guarantee computational productivity.
A convolutional-recurrent neural network serves as our main computational element to process sequential and spatial concrete mix data efficiently.
The model runs on a computing platform with powerful GPUs for its deployment.
The specific configuration serves as an essential requirement to make deep learning models perform quickly during training and inference, thus enabling fast processing of big datasets and intricate Using GPUs in the system enhances both processing speed and model scalability thus allowing the analysis of large datasets at high performance rates.
The development process utilizes TensorFlow and Keras frameworks as optimized deep learning applications specifically designed for this purpose.
The frameworks offer efficient neural network implementation, which includes automatic differentiation and GPU acceleration capabilities built right into their system.
The provided support maximizes resource usage in order to enhance model speed and accuracy during computations.
The computation process adopts standard software engineering principles that include logical modularization of code with optimized data arrays and parallel algorithm execution approaches.
Model maintenance becomes simpler through these practices, while calculations run faster, and the system becomes ready for growth requirements.
The predictive accuracy of our model was validated through statistical tests that included t-tests and ANOVA for comparing different configuration results.
The model performance was evaluated through an analysis that determined the effects of shifting GGBS or fly ash percentages.
Our model required this evaluation to demonstrate its performance across various concrete mixed conditions as we aimed to generate robust findings applicable to different production scenarios.
Int J Elec & Comp Eng.
Vol.
No.
October 2025: 4965-4982
Int J Elec & Comp Eng
ISSN: 2088-8708
The depiction in Figure 2 shows how the FLA and DBOA combine in a hybrid optimization structure without interruptions.
The diagram shows how the iterative steps of the dual algorithm function through parameter establishment and population initial creation.
The system computes fitness values to guide the search direction towards optimal solutions after this procedure.
The DBOA section allows the model to work without obstacles by readjusting individual positions through an equation designed for improved algorithm exploration.
By detecting local minimum obstacles, the trajectory makes intended modifications that allow smooth navigation towards optimal solutions.
The algorithm conducts a renewal procedure whenever it discovers an optimal solution, which allows it to perform more precise checks regarding boundary overstep from prior runs.
The FLA begins its hybrid system operations by establishing adjustable parameters that derive from DBOA outputs.
A dynamic feedback mechanism created between the algorithms improves both adaptability and robustness during the optimization process.
FLA separates its population across two groups, which conduct position updates based on fitness evaluations that are recalculated after every positional The algorithm divides its execution into multiple phases, which activate when iteration counts reach their defined thresholds in order to achieve efficient exploration and exploitation of the search space.
Both algorithms update their strategies with each iteration using changes in search space conditions to reach their final identification of the global best position and fitness.
This interactive algorithm mechanism both enhances individual algorithm effectiveness while utilizing their collective power to produce an optimized solution, which represents efficient management of exploration and exploitation resources needed for sophisticated optimization problems.
Our model evaluation includes examination of physics-informed machine learning (PIML) because we seek to elevate both interpretability and reliability of our predictive methods.
PIML strengthens model outcomes by implementing domain knowledge into training because it enables users to understand how concrete mix design principles affect predictions, which leads to more accurate, trustworthy results.
This method proves useful in material science because its analysis handles complex physical and chemical interactions that exhibit strong nonlinearity.
Dataset overview The main dataset used for this study examines concrete mixtures as a whole system to understand how components work together during compression tests.
The 1,030 samples have eight characteristics and one target variable.
The type .
and age are also listed.
With an average volume per mix of 17 kg, cement, the primary binder in concrete, is crucial to the structural behavior and longevity of the masonry industry.
Supplementary table cementitious materials average 73.
90 kg blast furnace slag and 54.
19kg fly ash.
Read Intro.
These improve durability and workability, but their proportions vary, providing a variety of experimental mixes.
Water, which is needed for concrete workability and strength, weighs 181.
57 kg per mix, whereas superplasticizers, which increase concrete fluidity without reducing strength, weigh 20 kilogram per mix.
The bulk of the mix is composed of both coarse and fine aggregates, weighing 92 kg and 773.
58 kg, respectively.
These factors affect concrete texture, density, and strength.
The average age of concrete samples was 45.
66 days, representing a cure period, a key parameter for strength growth layers.
Concrete undergoes chemical processes that release load-bearing properties.
The average compressive strength, or load-carrying ability, of concrete is 35.
82 MPa in this dataset.
Other highdimensional meta-features possess this scale, indicating aptitude for training predictive models that can handle many construction specifications and situations.
In conclusion, the dataset helps estimate quench-flow composite compressive strength by measuring ingredients and weather conditions and monitoring difficult concrete mix interactions.
This large data collection enables deeper examination of empirical relationships that affect concrete performance and customized mix designs for diverse construction applications.
In addition, we have included a comprehensive comparative Table 2 that details various optimization methods alongside the FLA and DBOA.
Each optimization technique's accuracy, together with its computational timing and benefits and drawbacks, appears in this table, which includes particle swarm optimization (PSO) and genetic algorithm, and Bayesian optimization.
An organized analysis enables better comprehension regarding why FLA and DBOA were selected for this research because of their particular advantages in resolving complex multi-dimensional optimization issues in concrete mixture design.
Our analysis of the dataset composition has been carried out to confirm the inclusion of diverse cement types across different geographic regions, which reduces biases related to materials and regions.
This analysis confirms our model's ability to apply to various cement materials through different testing systems for sustainable global construction deployment.
Enhancing concrete sustainability: a neural networks hybrid optimization A (EsraAoa Alhenaw.
A ISSN: 2088-8708 Table 2.
An overview of the concrete dataset's statistics Feature Cement Blast furnace slag Fly ash Water Superplasticizer Coarse aggregate Fine aggregate Age Strength (MP.
Count Mean Std Dev Min Max Preprocessing and exploratory data analysis In the initial stage of empirical research, the concrete dataset and accurate processing were The dataset was adjusted to scale all input features to similar ranges to generalize them and make neural network learning easier.
Standardizing variables prevents the model from prioritizing features with greater numbers, balancing feature relevance.
Figure 3 presents a comprehensive analysis of the distribution of various concrete mix components through histograms, each detailing the frequency and range of one particular ingredient.
The visualization begins with cement, displaying a right-skewed distribution that reflects a concentration of values at lower amounts with a gradual decline as the quantity increases.
This pattern suggests that smaller amounts of cement are more commonly used in the mixtures within the dataset.
Moving to the blast furnace slag and fly ash histograms, both show a significant number of samples containing minimal to no amounts, highlighted by the sharp peaks at the lower end of the scale.
These supplementary materials are applied on an optional basis or in minimal quantities across many concrete formulations found in the dataset.
The distribution shapes of water, along with superplasticizer and both coarse and fine aggregates, demonstrate typical usage patterns and value ranges within concrete mixtures.
The water distribution indicates that most concrete mixes use amounts that fall near their median values.
Superplasticizers are concentrated below the median amount, which shows many mixtures only use minimal amounts because this substance functions as an advanced additive rather than a regular mixture component.
Because of their essential role in the composition of concrete mixtures, coarse and fine aggregates appear to have a wide range of applications.
Standards in curing periods have shaped the observed peaks within the age distribution since many concrete mixtures receive their prescribed curing durations.
Compressive strength data follows a normal distribution, giving evidence of typical concrete mix behavioral patterns in the tested samples.
This visualization not only aids in understanding the typical properties of the materials used but also serves as a critical tool for identifying trends, anomalies, and the overall behavior of the components in concrete mix formulation.
Standardized exploratory data analysis (EDA) identified variable connections and distributions.
The histograms in Figure 3 show that the "distribution of distribution" can be of any kind and that the data distribution differs for each data set.
Because of low density at high cement concentration and denser locations at low cement levels, the cement feature displays a skewed right distribution.
This skewness is important because cement determines the strength and durability of concrete.
At zero, the distribution of fly ash and blast furnace slag is noteworthy.
thus, many mixtures do not use them.
As shown by their widespread use in various fields, they can improve concrete properties.
Water is crucial to the concrete mix ratio, but most components are balanced.
The right water levels affect curing, hydration, concrete strength, and movement.
Superplasticizer is less popular than Frequent Limestone.
Without water, mixed workability must be improved due to its increased frequency at lesser Given that the bulk of concrete's volume is composed of course and fine aggregate, their distributions aid in explaining this.
The variation in their quantities in the two samples suggests they can change mix density and strength.
Figure 4 correlation heatmap demonstrates a strong inverse relationship between water and superplasticizers, showing that superplasticizers reduce water use and strengthen mixes.
In the line plot as shown in Figure 5, the positive correlation between Age and Strength emphasizes the necessity of curing, where strength grows with time.
Concrete quality should be evaluated based on its age.
These exploratory discoveries help optimize concrete mix designs for greater performance and sustainability in construction by understanding the intricate interaction of concrete components.
This rigorous investigation illuminates concrete strength parameters, paving the road for building material science advancements.
Int J Elec & Comp Eng.
Vol.
No.
October 2025: 4965-4982
Int J Elec & Comp Eng
ISSN: 2088-8708
Figure 3.
Histograms of concrete components Figure 4.
Concrete features correlation matrix Enhancing concrete sustainability: a neural networks hybrid optimization A (EsraAoa Alhenaw.
A ISSN: 2088-8708 Figure 5.
Impact of features on compressive strength Hybrid convolutional-recurrent neural network (Conv1D-LSTM- GRU) architecture Convolutional and recurrent neural networks manage sequential input and predict concrete strength using composition profiles and time sequence in Table 3.
Layer one, a 1D convolutional layer, captures spatial relationships in the data.
Two kernels and 64 filters comprised this layer.
This reveals complicated patterns in sequential input data like concrete mix ingredient interactions.
The model's bidirectional long short-term memory (LSTM) follows the convolutional layer.
This layer has 50 LSTM units that can learn forward and backward dependencies.
Processing data in both directions reduces the need to comprehend each data point's context, revealing patterns that one-directional analysis misses.
Adding a second bidirectional layer with GRU increases model complexity and capability.
GRU
contains 50 units and accepts two-way data like LSTM.
GRUs are more effective and computationally Int J Elec & Comp Eng.
Vol.
No.
October 2025: 4965-4982
Int J Elec & Comp Eng
ISSN: 2088-8708
simpler, allowing faster training despite a modest performance difference.
LSTM's general features are reduced to the most important for the final choice by this layer.
The final layer is a thick network with one node for integrating characteristics and predicting concrete compressive strength.
In this layer, the model employs a linear activation function for regression issues using actual output.
The mean squared error loss function and Adam optimizer reduce training prediction errors, while backpropagation adjusts weights.
material science, this robust model is the best predictive analytical model as it captures data communication and general concrete strength features.
Table 3.
Model values and parameters Parameter Conv1D filters Conv1D Kernel size
Bidirectional LSTM units Bidirectional GRU units Dense layer units Optimizer Loss function
Total trainable parameters Value Adam
Mean squared error
91,893
FLA DBOA hybrid The hybrid optimization method uses the FLA and the DBOA to maximize the hyperparameters of neural network models.
This hybrid approach completely explores and utilizes the search space using both Step 1: Initialization The first step is to generate a population of possible solutions.
Every solution is represented by a vector of hyperparameters.
The initial population is created at random for each parameter within predetermined limitations.
ycuycn = ycuycn1 , ycuycn2 .
A ycuycnycu In this case, the ycn-th solution of the population is denoted by ycuycn , and the yc-th parameter of the ycn-th solution by ycuycnyc .
Step 2: Calculating fitness Determine the goal function's fitness for every solution.
The loss function of the neural network model is often the target function.
cuycn ) = yaycuycyc.
cuycn ) The neural network model's performance using the hyperparameters given by ycuycn is assessed by the fitness yce.
cuycn ).
Step 3: Firefly algorithm (FLA) mechanism The FLA component updates the population by simulating firefly activity.
Brighter .
solutions attract fireflies, and the path a firefly ycn follows to get close to another firefly yc is determined by .
= ycuycn yuyce Oeyuycycnyc .
cuyc Oe ycuycn ) yuyunycn where represents the attraction at yc = 0.
The symbol for the light absorption coefficient is yu.
Fireflies ya and ya are separated by ycycnyc .
The randomization parameter is denoted by .
The vector yunycn is random.
In order to prevent local optima, this equation introduces randomization while guaranteeing that each firefly travels in the direction of brighter fireflies.
Step 4: Dung beetle optimizer algorithm (DBOA) mechanism Through dung beetle simulation, the DBOA component refines the population.
The ideal solution affects the direction and step size of solution ycuycn :
= ycuycn cuycayceycyc Oe ycuycn ) yuyunycn where yu is an additional randomization parameter.
yuE is a scaling factor.
This approach ensures convergence towards an ideal solution by actively using the excellent areas that the firefly has found.
Enhancing concrete sustainability: a neural networks hybrid optimization A (EsraAoa Alhenaw.
A ISSN: 2088-8708 Step 5: Analyzing and choosing Evaluate the solutions produced by FLA and DBOA for fitness.
Update the optimal solution if the new one has a higher fitness value.
xbest = argmin f.
i ) Step 6: Iteration Continue until a predefined number of iterations is reached or convergence conditions are satisfied.
To arrive at the ideal neural network model hyperparameters, the solutions are improved over iterations.
EXPERIMENT RESULTS
Different models predict concrete compressive strength using the dataset, according to experiments as shown in Table 4.
Dung beetle optimizer (DBO), firefly algorithm (FLA), and hybrid (DBO FLA) optimized models are included.
The baseline model.
Conv1D-LSTM-GRU, has 49.
006192 MSE, 5.
MAE, 7.
000442 RMSE, and 0.
809815 R2.
These metrics quantify optimal model improvements.
The DBO model outperformed the baseline model with a test MSE of 44.
015296, reducing prediction error.
The DBO model had a smaller prediction error with a test MAE of 5.
Furthermore, a drop in the Test RMSE to 6.
634402 indicated that the forecasts were more accurate.
The test R2 increased to 829184, indicating a higher connection between the actual and projected compressive strengths.
As with FLA, performance improved.
It had 44.
748495 Test MSE, 5.
326784 MAE, 6.
RMSE, and 0.
826339 R2.
The FLA model outperformed the baseline model in prediction accuracy and error DBO FLA performed best in the experiment.
According to Test MSE, the model has the lowest prediction error, 40.
159906, and Test MAE was 5.
148660, which is the lowest prediction error.
Test RMSE reduced to 6.
337184, improving forecast accuracy.
With the highest test R2 of any model .
, the predicted and real compressive strengths showed the best connection.
According to the experiment, the hybrid DBO FLA model greatly enhances the ability to predict the compressive strength of concrete in Table 5.
The DBO FLA hybrid, in particular, is one of the optimum types, showing how modern optimization methods refine neural network hyperparameters, improving material science predictions.
Comparing our results to related works, because of the size of our dataset, error measurements like root mean squared error (RMSE) and mean squared error (MSE) may rise.
Bigger datasets are more complicated and varied, which makes prediction more difficult and raises absolute error levels.
However, our hybrid optimization strategy is robust and effective because our models improve consistently.
Table 4.
Results of the forecasting test for concrete compressive strength
Model
Conv1D-LSTM-GRU
DBO
FLA
DBO FLA
Test MSE Test MAE Test RMSE Test R2 Table 5.
Comparing the outcomes of the experiment with related works Ref.
AdaBoost .
Boosting .
Bagging .
model_1 Our Work
DBO Our Work
FLA Our Work
DBO FLA Our Work
RMSE
MSE
MAE
MAPE
The superiority of the developed models is underscored by the integration of the FLA and DBOA with a convolutional-recurrent neural network, offering significant enhancements over traditional modeling This hybrid optimization approach combines the accuracy of DBOA's local search with FLA's global search power, allowing the model to more successfully traverse intricate optimization landscapes and steer clear of local minima.
Enhanced accuracy happens when predicting concrete compressive strength because of this method, which is vital for reliable construction material assessment.
Int J Elec & Comp Eng.
Vol.
No.
October 2025: 4965-4982
Int J Elec & Comp Eng
ISSN: 2088-8708
The convolutional-recurrent architecture successfully detects spatial and temporal dependencies that naturally exist in concrete mix databases.
Organizations depend on this functionality to understand complex interactions that impact the material properties of concrete components.
The model demonstrates robustness through successful adoption across different datasets, which enables performance maintenance across various operational conditions.
The developed models benefit from efficient operation since GPU acceleration supports the preservation of high computational efficiency, together with accurate prediction speed.
The system delivers outstanding benefits to industries that require rapid resource management and quick execution times.
The adoption of SHAP among feature analysis methods brings both enhanced model interpretability as well as transparency.
The model gains wider practical use because users and decision-makers develop trust through feature explanation while gaining comprehension of which input elements lead to specific output The extensive understanding of model prediction reasons stands equally important to prediction accuracy, thus making model transparency an essential matter for specific sectors.
When combined, these architectural elements demonstrate a major improvement in utilizing machine learning algorithms to forecast concrete strength, which results in outstanding operational effectiveness for practical and industrial use.
CONCLUSION
The project investigates methods of reducing CO2 emissions from the construction sector by enhancing concrete properties through the use of supplementary cementitious materials (SCM.
, such as ground granulated blast furnace slag (GGBS) and fly ash (FA).
The prediction of concrete compressive strength has increased thanks to thorough data analysis and potent machine learning techniques.
The hybrid optimization method fine-tuned Convolutional-Recurrent Neural Network hyperparameters using the FLA and DBOA.
It optimized the model better than earlier methods.
Of all the models tested, the hybrid model scored the lowest MSE.
RMSE, and R2.
Our results show that FLA and DBOA work well for accuracy in material science forecasting.
The hybrid method fully uses the hyperparameter search space for more accurate predictions.
SCMs in concrete mixtures benefit the building industry's low-carbon circular economy.
Finally, this research offers a new way to optimize concrete mix designs, adding to sustainable construction The results demonstrate that advanced optimization and SCMs may enhance material performance and environmental sustainability.
To enhance them, future studies may apply these strategies to more concrete manufacturing processes and incorporate further optimization approaches.
FUNDING INFORMATION
Authors state no funding involved.
AUTHOR CONTRIBUTIONS STATEMENT
This journal uses the Contributor Roles Taxonomy (CRediT) to recognize individual author contributions, reduce authorship disputes, and facilitate collaboration.
Name of Author EsraAoa Alhenawi Ayat Mahmoud AlHinawi Zaher Salah Omar Alidmat Esraa Abu Elsoud Raed Alazaidah Bashar Rizik AlSayyed C : Conceptualization M : Methodology So : Software Va : Validation Fo : Formal analysis ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue I : Investigation R : Resources D : Data Curation O : Writing - Original Draft E : Writing - Review & Editing ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue Vi : Visualization Su : Supervision P : Project administration Fu : Funding acquisition CONFLICT OF INTEREST STATEMENT Authors state no conflict of interest.
Enhancing concrete sustainability: a neural networks hybrid optimization A (EsraAoa Alhenaw.
A ISSN: 2088-8708
DATA AVAILABILITY
The data that support the findings of this study are available from the corresponding author.
ZS, upon reasonable request.
REFERENCES