Sustainable construction has become a global priority as the demand for infrastructure grows alongside escalating environmental concerns. Recycled coarse aggregate (RCA), obtained from construction and demolition waste, offers a practical and eco-efficient alternative to natural aggregates. Its use directly contributes to reduced landfill disposal, lower carbon emissions, and conservation of natural resources. However, despite these benefits, RCA possesses several limitations—high porosity, adhered mortar, microcracking, and weaker interfacial transition zones (ITZ)— which collectively diminish the mechanical strength and durability of concrete produced with higher RCA content. These drawbacks restrict the structural application of RCA concrete and highlight the need for performance- enhancing strategies.Recent advances in nanotechnology have introduced graphene oxide (GO) as a highly promising additive for cementitious systems. GO exhibits a two-dimensional nanosheet structure with abundant oxygen functional groups, enabling strong surface reactivity and improved bonding with calcium–silicate–hydrate (C–S–H) gel. Numerous studies report that GO promotes nucleation of hydration products, refines pore structure, enhances ITZ density, and improves mechanical performance even at very low dosages. This creates a unique opportunity to offset the weaknesses of RCA and produce high-performance sustainable concrete.

Parallel to material innovations, machine learning (ML) has emerged as a powerful analytical tool capable of accurately predicting concrete properties based on limited experimental data. Modern ensemble algorithms such as XGBoost, LightGBM, and CatBoost excel at modelling nonlinear and multi-variable relationships, making them well-suited for predicting RCA–GO concrete behaviour. ML not only strengthens result validation but also facilitates rapid mix optimisation, reducing dependence on extensive laboratory testing.

In this context, the present study explores the combined effect of GO modification and ML-based prediction on the performance of M40 concrete containing 20% RCA. The research aims to (i) investigate the mechanical and durability characteristics of RCA–GO concrete, (ii) identify the optimum GO dosage, and (iii) develop accurate ML

models to validate and forecast concrete performance. The novelty of this work lies in integrating nano-enhancement with data-driven modelling to establish a modern framework for designing sustainable, high-performance concrete mixtures.

The use of recycled coarse aggregate (RCA) has gained significant attention as the construction industry moves toward sustainability. Numerous studies between 2018 and 2025 consistently report that RCA exhibits higher porosity, adhered mortar, and microcracking compared to natural aggregates, which collectively weaken the mechanical performance of concrete. Researchers have observed that compressive strength typically decreases with increasing RCA replacement due to reduced aggregate stiffness and weaker interfacial transition zones (ITZ).

Despite these drawbacks, many authors highlight that the environmental and economic benefits of RCA justify its use in structural and pavement applications when its limitations are properly addressed through material modification.

Graphene oxide (GO) has emerged as one of the most effective nano-modifiers capable of improving the microstructure of cementitious composites. Studies conducted after 2019 demonstrate that GO enhances hydration kinetics, fills microvoids, and promotes the formation of dense C–S–H gel, resulting in significant improvement in mechanical properties. Researchers have reported increases in compressive and tensile strength in the range of 5– 20% at optimal dosages between 0.02% and 0.06% by weight of cement. Beyond the optimum range, several works note that GO tends to agglomerate, reducing its efficiency and adversely affecting workability and strength. Recent investigations also emphasize GO’s ability to refine pore structure, decrease capillary suction, and enhance resistance against chloride ingress and acid attack, making it a promising additive for durability enhancement.

The combined use of RCA and GO has gained momentum in recent research, as GO’s microstructural benefits can offset weaknesses inherent in recycled aggregates. Studies from 2021 onward demonstrate that GO improves the weak ITZ region of RCA concrete, resulting in better mechanical performance even at higher RCA replacement levels. Researchers have observed that GO strengthens the bonding between adhered mortar and the surrounding paste, which is typically the weakest region in RCA-based mixtures. Furthermore, GO-modified RCA concrete displays reduced water absorption and improved resistance to sulphuric and hydrochloric acid exposure, highlighting GO’s potential to restore the durability deficits associated with recycled aggregates.

Parallel to material advancements, machine learning (ML) has become an increasingly valuable tool for concrete property prediction. Between 2020 and 2025, ensemble algorithms such as XGBoost, LightGBM, and CatBoost have demonstrated high prediction accuracy for compressive strength, tensile capacity, and durability parameters of conventional and recycled aggregate concrete. These models are particularly effective in capturing nonlinear interactions between mix constituents, curing conditions, and performance outcomes. Several researchers have incorporated ML in sustainable concrete studies to validate experimental results and to reduce laboratory workload by enabling rapid prediction of multi-parameter systems. CatBoost, in particular, has been repeatedly highlighted for its superior performance on small and structured experimental datasets, making it suitable for concrete mix design research.

Recent literature further emphasizes the value of combining nano-modification with ML-based prediction. Works published after 2023 show that ML-assisted approaches significantly improve the reliability of performance forecasting for nano-engineered concretes, allowing researchers to identify optimum nano-admixture dosages, assess parameter influence, and evaluate material stability under various environmental conditions. These studies indicate that integrating GO modification with ML modelling provides a comprehensive approach for analyzing mechanical and durability improvements in RCA concrete.

Despite notable advancements, the literature reveals two clear gaps. First, only limited research has examined the combined effect of 20% RCA with varying GO dosages under both mechanical and durability frameworks. Second, the integration of GO-enhanced RCA concrete with advanced ML validation—especially using CatBoost and ensemble averaging—remains scarce. These gaps highlight the need for a detailed experimental study supported by robust ML prediction to fully understand the performance potential of RCA–GO concrete for structural and rigid pavement applications.

The experimental programme was designed to evaluate the mechanical and durability performance of M40-grade concrete containing 20% recycled coarse aggregate and varying dosages of graphene oxide. A total of six concrete mixes were prepared, each with a fixed RCA content of 20% and GO dosages of 0%, 0.02%, 0.04%, 0.06%, 0.08%, and 0.10%. All mixes were proportioned using a constant water–cement ratio of 0.40 and produced under identical batching, mixing, and curing conditions to ensure comparability. The primary objective of the experimental programme was to identify the optimum GO dosage capable of compensating for the inherent weaknesses of RCA, while simultaneously generating a consistent dataset suitable for machine learning validation.

The mixing procedure followed a controlled and repeatable sequence to ensure uniform distribution of all constituents. Initially, cement, fine aggregate, natural aggregate, and saturated surface-dry RCA were dry-mixed to achieve a homogeneous base. The pre-dispersed GO suspension was then introduced gradually with a portion of the mixing water, followed by the remaining water containing the superplasticizer. The mixture was blended thoroughly to avoid GO agglomeration and to maintain a workable and cohesive mix. Slump measurements were taken immediately after mixing to monitor the influence of RCA and GO on workability.

Fresh concrete was placed into moulds in two layers and compacted using a table vibrator to eliminate air voids. For mechanical testing, cubes (150 × 150 × 150 mm), cylinders (100 × 200 mm), and beams (100 × 100 × 500 mm) were cast. Durability tests required separately prepared specimens of appropriate dimensions as per relevant standards. All specimens were demoulded after 24 hours and transferred to a curing tank maintained at 27 ± 2°C. Testing was carried out at curing ages of 7, 28, 56, and 90 days to capture both early-age and long-term performance characteristics.

Mechanical tests included compressive strength, split tensile strength, flexural strength, and static modulus of elasticity. Compressive strength tests were performed according to IS 516 and ASTM C39, while split tensile and flexural strength tests adhered to IS 5816 and IS 516 protocols, respectively. The modulus of elasticity was determined using the stress–strain response of cylinder specimens. Durability assessments focused on water absorption, sorptivity, acid resistance under sulphuric and hydrochloric acid exposure, rapid chloride penetration testing (RCPT), and ultrasonic pulse velocity (UPV). Each test was conducted in accordance with established IS and ASTM procedures to ensure reliability and reproducibility.

All experimental data were systematically recorded and organized into structured datasets for subsequent machine learning analysis. The performance of each mix was evaluated holistically by integrating mechanical, durability, and

microstructural responses, thereby enabling a comprehensive understanding of the influence of GO on RCA- modified concrete. The resulting datasets formed the foundation for developing predictive models using XGBoost, LightGBM, and CatBoost, allowing experimental observations to be validated through advanced data-driven methods.

The machine learning (ML) framework was developed to validate and predict the mechanical and durability behaviour of M40-grade concrete incorporating 20% recycled coarse aggregate and graphene oxide as a nano- modifier. The experimental dataset served as the basis for training, testing, and evaluating three state-of-the-art gradient-boosting ensemble models—XGBoost, LightGBM, and CatBoost. These models were selected because of their proven efficiency in handling nonlinear relationships, small to medium-sized datasets, and complex multi- parameter interactions common in concrete materials research.

The mechanical properties demonstrated a consistent improvement with increasing GO dosage up to 0.04–0.06%. The RCA20 concrete without GO showed lower strength due to weak interfacial bonding, but GO nanosheets significantly enhanced hydration rates, ITZ density, and crack-bridging capability. Peak values occurred at 0.06% GO, where compressive strength increased by ~ 7.5% and flexural strength by ~ 17% compared to the unmodified RCA20 mix. Beyond 0.06%, agglomeration of GO reduced its effectiveness and caused slight performance decline.

Durability results follow the same nonlinear trend as mechanical properties.

GO at 0.04–0.06% significantly refined pore structure, reducing water absorption by up to 17% and RCPT by ~ 24%. Sorptivity declined steadily due to enhanced microstructural densification. Acid resistance improved as mass loss decreased, supported by higher UPV values indicating stronger internal integrity. The decline at higher GO dosages is attributed to nanosheet clustering, which creates weak zones in the matrix.

All three models captured the nonlinear behaviour of GO dosage extremely well. CatBoost delivered the highest accuracy (R² up to 0.971), due to its ability to model small datasets and complex interactions. The ML predictions closely matched experimental peaks at 0.04–0.06% GO and correctly predicted strength decline at higher dosages.

This study demonstrates that the combined use of recycled coarse aggregate and graphene oxide provides an effective pathway for developing high-performance and sustainable M40-grade concrete. Although the inherent weaknesses of RCA, including adhered mortar and high porosity, tend to reduce strength and durability, the incorporation of graphene oxide successfully counteracts these limitations by refining the microstructure and strengthening the interfacial transition zone. Experimental results confirm that GO significantly enhances mechanical performance and durability up to an optimum dosage of 0.04–0.06%, where notable improvements in compressive strength, tensile capacity, flexural response, pore refinement, and resistance to permeability and acid attack were observed. Beyond this range, performance declined slightly due to agglomeration of nanosheets, emphasizing the importance of dosage control.

The machine learning models developed in this study—XGBoost, LightGBM, and CatBoost—validated the experimental outcomes with high predictive accuracy. CatBoost consistently achieved the best results, with R² values exceeding 0.95 for both mechanical and durability parameters, reflecting its ability to model complex nonlinear interactions within the RCA–GO system. The strong alignment between ML predictions and experimental measurements demonstrates that data-driven modelling can reliably capture the behaviour of nano-modified recycled aggregate concrete. These results highlight the value of ML for rapid performance prediction, mix optimization, and reduced reliance on intensive laboratory testing.

Overall, the integration of graphene oxide with RCA, supported by advanced machine learning validation, offers a modern and efficient framework for designing high-performance sustainable concrete. The findings establish that RCA20 concrete, when optimally modified with GO, can achieve mechanical and durability performance comparable to or exceeding that of conventional concrete. This research contributes a robust experimental and computational foundation for future development of eco-efficient concrete for structural and rigid pavement applications.