The study focuses on the bearing strength of subgrade soils strengthened with geosynthetic materials, which are used in road construction to improve weak soils through functions including separation, reinforcement, filtration and drainage. CBR is a key measure of subgrade bearing capacity and informs pavement design. Conventional soaked CBR testing can be slow and labor-intensive: specimens remain submerged for 96 hours, while Proctor compaction tests to determine optimum moisture content and maximum dry density can extend the work on an individual sample to five or six days. Heterogeneous site conditions may require testing many samples. The compiled dataset therefore offers a potential route to estimating CBR from properties that can be obtained more readily, although the study does not establish that the models can replace laboratory testing in routine practice.

To build the database, the researchers combined 90 soaked CBR observations from nine previously published studies. The records covered sandy and fine-grained soils classified under the Unified Soil Classification System, with samples representing several combinations of soil properties, geosynthetic materials and reinforcement configurations. Eleven predictors were used: liquid limit, plastic limit, plasticity index, maximum dry unit weight, optimum moisture content, percentage of fines, percentage of sand, geosynthetic tensile strength, number of reinforcement layers, position of the first layer and position of subsequent layers. Soaked CBR was the target variable. Across the dataset, CBR ranged from 1.32% to 88.3%, with a mean of 16.067%. Reinforcement tensile strength ranged from 0.75 to 140, and the number of layers ranged from one to three.

The model comparison used a common 80/20 partition, retaining 72 records for training and holding out 18 for final testing. Input scaling parameters were derived only from the training data and then applied unchanged to the test data, limiting information leakage. The authors evaluated plain and AOA-assisted versions of three learners under the same split: ANN, Ridge regression and XGBoost. AOA is a population-based metaheuristic inspired by the hunting behavior of Aquila eagles. It was used to tune model weights and hyperparameters, with mean squared error serving as the optimization objective for the neural-network training. The AOA configuration used a population of 150 and up to 500 generations, and the ANN had an 11–6–1 architecture: 11 inputs, six hidden neurons and one output neuron.

AOA improved the reported test performance of every base model, although the size of the gain differed. For ANN, R² increased from 0.929 to 0.945, MSE declined from 3.06 to 2.87 and MAE fell from 18.09 to 13.87. Ridge remained the weakest performer despite optimization: AOA-Ridge reached R² = 0.831, MSE = 4.90 and MAE = 42.81, compared with R² = 0.815, MSE = 5.17 and MAE = 47.09 for plain Ridge. The AOA-XGBoost model ranked first under the study’s weighted color-intensity scheme, which assigned weights of 0.5 to R², 0.3 to MSE and 0.2 to MAE. Its advantage was especially visible in MAE, which fell by 57.1% relative to plain XGBoost. For this dataset and modeling setup, AOA-assisted XGBoost was the most accurate of the six evaluated models.

Five-fold cross-validation on the training data was used to test whether the ranking depended on the particular hold-out split. AOA-XGBoost again performed best, with cross-validated R² = 0.8819 ± 0.0765, root mean squared error of 6.4218 ± 3.1846, MAE of 3.5217 ± 1.4826, mean absolute percentage error of 19.6842 ± 4.7315%, and A20 = 0.6427 ± 0.1186. The A20 metric represents the proportion of predictions falling within ±20% of the corresponding measured values. AOA-ANN ranked second, with R² = 0.8347 ± 0.1062, while AOA-Ridge had R² = 0.6487 ± 0.3628. The broadly consistent ordering between the held-out test and cross-validation results supports the reported advantage of the optimized tree-based model, but the cross-validated R² is substantially lower than the 0.980 obtained on the test set, illustrating the sensitivity of performance estimates in a small dataset.

Training behavior for the AOA-ANN was also monitored over 500 generations. Its mean squared error fell from approximately 177 initially to around 95 in the first 20 epochs, then declined in stages to approximately 58 at epoch 40, 50 near epoch 160 and 44 around epoch 190. Improvement slowed after roughly 200 epochs, reaching about 40 by epoch 250 and approximately 38 at the end. The authors interpreted this as convergence toward a near-stable solution, with further optimization offering limited gains. The study also published the final ANN weights and biases and provided a worked calculation for a non-plastic reinforced soil condition, yielding a predicted CBR of 41.35%. That example lies within the observed CBR range and was presented as an implementation demonstration, not as independent validation of the model outside the compiled data.

Two sensitivity methods, the Cosine Amplitude Method (CAM) and Garson’s algorithm, were used to estimate the relative influence of the 11 inputs. Their combined analysis assigned 29.4% of the sensitivity to the number of reinforcement layers and 21.8% to geosynthetic tensile strength. The position of the first reinforcement layer contributed 12.1%, followed by the position of subsequent layers at 9.9% and plastic limit at 7.2%; the other variables together accounted for 19.5%. Thus, the four reinforcement-related variables collectively represented more than 73% of the model’s reported sensitivity. CAM and Garson ranked some secondary variables differently, however, and a separate stability analysis gave dry unit weight the most consistent rank even as the two methods disagreed about the ordering of tensile strength, sand and fines. The percentages are model-based importance measures rather than estimates of causal effects or direct physical contributions to strength.

The most important limitation is the database’s size and composition. Only 90 records were assembled, and differences among source studies could not be eliminated in soil type, material characteristics, specimen preparation or testing conditions. Five-fold validation and the independent hold-out set reduce but do not remove the risk of overfitting, especially for a model with 11 inputs and relatively few observations. The models should therefore be used primarily within the parameter domain represented by the compiled data; predictions outside the reported ranges involve extrapolation. The neural network and XGBoost also retain elements of a black box, and the sensitivity results do not replace direct testing or mechanistic evaluation of soil–reinforcement interaction. The study’s strongest conclusion is therefore bounded but promising: within its experimental dataset and evaluation protocol, AOA-XGBoost was the most accurate and consistently top-ranked model for estimating soaked CBR, and larger independent validation will be needed before its performance can be established beyond the current scope.

Subject of Research: Optimization-assisted machine learning for predicting the California Bearing Ratio of geosynthetic-reinforced subgrade soil

Article Title: New hybrid models integrating the aquila optimizer with machine learning algorithms to predict the strength of geosynthetic-reinforced subgrade soil

Article References: Tiang, J.-J., Ateş, B., Eirgash, M. A., Zhao, Y., Anand, P., Sharma, A., & Lim, W. H. (2026). New hybrid models integrating the aquila optimizer with machine learning algorithms to predict the strength of geosynthetic-reinforced subgrade soil. Frontiers in Built Environment, 12, Article 1947550. https://doi.org/10.3389/fbuil.2026.1947550

Image Credits: AI Generated

DOI: 10.3389/fbuil.2026.1947550

Keywords: geotechnical engineering, machine learning, XGBoost, metaheuristic optimization, California Bearing Ratio, geosynthetics, subgrade soil, sensitivity analysis, pavement design, hybrid models, hybrid, models

The review, published in Discover Geoscience, is not a typical narrative survey. The team followed the PRISMA 2020 guidelines for systematic reviews, combing databases including Scopus, Web of Science, ScienceDirect, IEEE Xplore, PubMed, Google Scholar and ResearchGate for papers published between 2010 and 2025. Boolean search strings combined terms such as “soil stabilization” and “ground improvement” with “machine learning,” “deep learning” and “artificial intelligence,” and then filtered again for “green materials,” “industrial waste” and “circular economy.” The initial sweep captured 685 publications. After removing 214 duplicates, screening 471 records by title and abstract, and excluding 77 articles that lacked sustainable binders, predictive models or sufficient experimental data, the researchers were left with 111 primary studies that form the evidentiary backbone of the analysis.

To map the intellectual geography of the field, the authors turned to VOSviewer version 1.6.19, a bibliometric tool that visualizes keyword co-occurrence, author collaboration and institutional networks. Keywords appearing fewer than five times were filtered out of the co-occurrence analysis, and co-authorship networks required a minimum of three documents per author. The resulting maps reveal a discipline that has moved through three distinct chronological phases. In the foundational period before 2015, researchers applied relatively simple artificial neural networks to predict classic geotechnical quantities, chiefly the Unconfined Compressive Strength, or UCS, of stabilized soil specimens and the California Bearing Ratio, or CBR, used in pavement design. A middle phase integrated sustainability objectives, as algorithms such as Support Vector Machines and Random Forests were trained on mixes containing fly ash, biochar and geopolymer binders, aligning the field explicitly with circular economy thinking. The current frontier, the brightest hotspot in the density maps, extends into holistic soil health, with models probing soil microbiology, hydrogen-ion concentration and pollutant behavior.

The geographic story told by the publication data is equally striking. The analysis projects exponential growth led by several countries working largely on parallel tracks. China maintained the highest overall publication volume through 2023 and is projected to reach roughly 95 papers by 2025. India is expected to overtake it, with an estimated 100 publications by 2025, while Iran, anchored by institutions such as Islamic Azad University, is projected at around 85. Vietnam shows the fastest exponential trajectory, climbing from minimal output to an estimated 75 publications driven by AI-based strength prediction research. Turkey, applying machine learning to ground improvement in seismically active regions, is projected at 60 publications, with Malaysia at about 45 and Australia, emphasizing high-impact reviews and predictive modeling of soil-structure interaction, at roughly 35. The review notes that emerging clusters of prolific post-2020 authors signal where the next wave of the field is likely to concentrate.

What makes machine learning so attractive to geotechnical engineers is the stubbornly nonlinear character of the problem itself. When a sustainable additive is blended into soil, the mechanical response emerges from complicated chemical and physical interactions: pozzolanic reactions, cation exchange, particle packing, moisture redistribution and curing chemistry all couple together in ways that resist simple closed-form equations. Traditional characterization relies on laboratory tests such as UCS and CBR that are resource-intensive, labor-hungry and slow, forcing engineers to run many trial mixes before settling on a design. Data-driven models invert that workflow. Once trained on published experimental results, a model can estimate the strength or bearing capacity of a proposed soil-additive blend in seconds, letting researchers screen candidate mixes computationally before committing bench time and materials to the most promising candidates.

The review’s synthesis of the underlying materials science adds practical texture to the computational story. Rice husk ash, one of the most heavily experimented additives, is rich in silica at roughly 70 to 90 percent, and when burned under controlled conditions it forms amorphous silica that reacts pozzolanically with calcium compounds in soil. The compiled data show that increasing rice husk ash content lowers the liquid limit and plastic limit of treated soil while raising UCS and CBR values, with maximum dry density following the same beneficial trend. Fly ash, the fine powder captured from the flue gases of coal-fired power plants, contributes silica, alumina and calcium oxide; the review identifies an optimal dosage window of roughly 15 to 25 percent for peak CBR performance. Shredded plastic waste, being non-biodegradable, must be dosed cautiously, with maximum CBR gains occurring at only 1 to 2 percent content. Aluminum slag tolerates much higher concentrations, reaching maximum compressive strength at an optimal content near 15 percent.

On the algorithmic side, the verdict from the assembled literature is that three families of models currently dominate in accuracy: artificial neural networks, support vector machines and random forests. The authors report that ensemble and deep learning algorithms have successfully predicted laboratory datasets with coefficients of determination between 0.88 and 0.96, a level of agreement that would have been difficult to achieve with classical empirical correlations. Those numbers come with an important asterisk, however. Model performance is heavily dependent on data quality, hyperparameter optimization and how well the input features represent the underlying soil-additive chemistry. A model trained on one site’s laterite may transfer poorly to another region’s soft marine clay, a generalization problem that haunts much of the published literature.

Indeed, the review is unusually candid about the field’s weaknesses. Most existing models have been built on small, site-specific laboratory datasets, which restricts their applicability across the enormous diversity of natural soils. Few studies have evaluated the long-term environmental durability of treated ground, such as performance through freeze-thaw cycles or the leaching behavior of additives in wet conditions, and almost none have integrated real-time sensor data from actual field installations. There is also an interpretability problem: the black-box character of deep neural networks means that even highly accurate predictions may offer little physical insight into the chemical interactions between a soil and its additive, leaving engineers to trust correlations they cannot fully explain. The authors flag these gaps, alongside the sheer scarcity of shared training data, as the critical bottlenecks between today’s promising laboratory results and tomorrow’s field-scale adoption.

The broader stakes extend well beyond geotechnical journals. The construction industry is under intensifying pressure to reduce its dependence on conventional cement and lime binders because of their outsized contribution to environmental pollution, and the review makes clear that greener substitutes alone are not the answer, since natural fibers and industrial byproducts frequently behave inconsistently across soil types and curing regimes. Pairing those materials with data-driven prediction is what converts them from experimental curiosities into design-ready tools, letting engineers quantify trade-offs between strength, cost and carbon before a single cubic meter of ground is treated. The authors frame this convergence of information science and sustainability as a genuine paradigm shift, one that connects soil stabilization to circular economy goals by giving waste streams such as fly ash, slag, rice husk ash and even plastics a quantified second life as engineering materials.

For a field that has quietly supported nearly every major infrastructure project of the modern era, soil stabilization has rarely attracted headlines. That may be changing. A discipline once defined by trial mixes and look-up charts is now publishing keyword co-occurrence networks, temporal overlay maps and national publication projections that read like the output of a thriving computational science. The Bangladesh-led review offers a clear summary of fifteen years of progress under strict systematic standards, and its message to practitioners is direct: the algorithms are ready, the green additives are ready, but the databases, field validation studies and explainable models needed to bridge laboratory and ground truth are not yet in place. Filling that gap, the authors argue, will determine whether the next generation of foundations is both smarter and substantially cleaner than the last.

Subject of Research: Sustainable soil stabilization using machine learning and deep learning

Article Title: A systematic review on sustainable soil stabilization using machine learning and deep learning

Article References: Islam, N., Hawladar, M. N., Rahman, S., Uddin, A., Shehab, M. F. H., Pramanik, M. R., Islam, I., & Haque, M. F. (2026). A systematic review on sustainable soil stabilization using machine learning and deep learning. Discover Geoscience, 4(1), Article 352. https://doi.org/10.1007/s44288-026-00723-y

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00723-y

Keywords: soil stabilization, machine learning, deep learning, geotechnical engineering, sustainable materials, fly ash, rice husk ash, circular economy, unconfined compressive strength, California Bearing Ratio, bibliometric review, green construction