STUDY ON MACHINE LEARNING METHODS FOR COMPRESSIVE STRENGTH PREDICTION USING CONCRETE MANUFACTURING CONTROL DATA

Development of compressive strength prediction platform for concrete materials based on machine learning techniques

Ultra‑High‑Performance Concrete (UHPC) is a considerably advanced cementitious concrete with great characteristics of strength and durability, but the compressive strength is highly dependent on the multi‑faceted interplay between mixture proportions and curing conditions. These interactions are nonlinear and multivariate, making it difficult to accurately estimate the UHPC compressive strength using the previous experimental and empirical methods. In the paper, a Random Forest (RF) regression model has been constructed to estimate UHPC compressive strength based on a large‑scale dataset of 810 samples and 13 predictors (material composition and curing parameters). Multiple statistical measures were strictly used to evaluate the performance of the model, such as R2, RMSE, MAE, MAPE, and CVRMSE, as well as 10‑fold cross‑validation to evaluate stability and ability to generalize. The optimized RF model had a high predictive accuracy with a value of 0.96 on the testing set and small values of errors, which showed high robustness and consistency in diverse segmentations of data. Hyperparameter tuning also improved the model performance by finding a balance between model complexity and generalization. SHAP (Shapley Additive Explanations) analysis was used to enhance the transparency and interpretability of the models, to measure the contribution of the individual input feature to the compressive strength predictions. The findings demonstrated that curing age, fibre, silica fume, and dosage of superplasticizer were the most significant parameters that controlled the strength development of UHPC. The suggested modeling framework reveals the efficiency of bringing ensemble machine learning along with explainable artificial intelligence methods to provide accurate, reliable, and interpretable predictions of UHPC compressive strength, which creates a useful instrument in the process of mix design optimization and performance evaluation.

Poor sleep quality amplifies pain experience. However, it remains uncertain regarding whether subjectively or physiologically assessed sleep parameters better predict pain, given reported discrepancies between the subjective and physiological measures of sleep quality. This secondary analysis examined data from 556 adults in the Cleveland Family Study. Clinical pain severity was assessed using the two-item SF-36 pain subscale. Subjective sleep quality was measured by the Functional Outcomes of Sleep Questionnaire, and physiological sleep quality was assessed via polysomnography. To examine complex sleep-pain relationships, machine learning (ML) models were utilized. The dataset was split into training (70%) and testing (30%) subsets. After feature selection, Gradient Boosted Regression (GBR) was applied to develop a predictive model, validated by Random Forest (RF) and Elastic Net Regression (ENR). Results indicated similar prediction accuracy and feature importance across GBR, RF, and ENR. Sociodemographic, clinical, and anthropometric factors explained 32% of the variance in clinical pain using GBR; inclusion of sleep profiles increased explained variance to 35%. Analysis of the feature importance revealed subjective experiences of post-sleep activity and vigilance as the strongest predictors of pain severity, rather than physiologically measured sleep quality. We further confirm this finding using K-means clustering analysis. These findings highlight the significant role of subjective sleep quality compared to physiological sleep measures in influencing clinical pain severity. They also suggest the potential value of further investigating whether interventions aimed at improving perceived sleep quality may more effectively improve pain outcomes.

With the sustainable development of the construction industry, recycled aggregate (RA) has been widely used in concrete preparation to reduce the environmental impact of construction waste. Compressive strength is an essential measure of the performance of recycled aggregate concrete (RAC). In order to understand the correspondence between relevant factors and the compressive strength of recycled concrete and accurately predict the compressive strength of RAC, this paper establishes a model for predicting the compressive strength of RAC using machine learning and hyperparameter optimization techniques. RAC experimental data from published literature as the dataset, extreme gradient boosting (XGBoost), random forest (RF), K-nearest neighbour (KNN), support vector machine regression Support Vector Regression (SVR), and gradient boosted decision tree (GBDT) RAC compressive strength prediction models were developed. The models were validated and compared using correlation coefficients (R2), Root Mean Square Error (RMSE), mean absolute error (MAE), and the gap between the experimental results of the predicted outcomes. In particular, The effects of different hyperparameter optimization techniques (Grid search, Random search, Bayesian optimization-Tree-structured Parzen Estimator, Bayesian optimization- Gaussian Process Regression) on model prediction efficiency and prediction accuracy were investigated. The results show that the optimal combination of hyperparameters can be searched in the shortest time using the Bayesian optimization algorithm based on TPE (Tree-structured Parzen Estimator); the BO-TPE-GBDT RAC compressive strength prediction model has higher prediction accuracy and generalisation ability. This high-performance compressive strength prediction model provides a basis for RAC’s research and practice and a new way to predict the performance of RAC.

Abstract. Firn density plays a crucial role in assessing the surface mass balance of the Antarctic ice sheet. However, our understanding of the spatial and temporal variations in firn density is limited due to (i) spatial and temporal limitations of in situ measurements, (ii) potential modelling uncertainties, and (iii) lack of firn density products driven by satellite remote-sensing data. To address this gap, this paper explores the potential of satellite microwave radiometer (Special Sensor Microwave Imager/Sounder (SSMIS)) and scatterometer (Advanced Scatterometer (ASCAT)) observations for assessing spatial and temporal dynamics of dry-firn density over the Antarctic ice sheet. Our analysis demonstrates a clear relation between density anomalies at a depth of 40 cm and fluctuations in satellite observations. However, a linear relationship with individual satellite observations is insufficient to explain the spatial and temporal variation in snow density. Hence, we investigate the potential of a non-linear random forest (RF) machine learning approach trained on radiometer and scatterometer data to derive the spatial and temporal variations in dry-firn density. In the estimation process, 10 years of SSMIS observations (brightness temperature) and ASCAT observations (backscatter intensity) is used as input features to a random forest (RF) regressor. The regressor is first trained on time series of modelled density and satellite observations at randomly sampled pixels and then applied to estimate densities in dry-firn areas across Antarctica. The RF results reveal a strong agreement between the spatial patterns estimated by the RF regressor and the modelled densities. The estimated densities exhibit an error of ±10 kg m−3 in the interior of the ice sheet and ±35 kg m−3 towards the ocean. However, the temporal patterns show some discrepancies, as the RF regressor tends to overestimate summer densities, except for high-elevation regions in East Antarctica and specific areas in West Antarctica. These errors may be attributed to underestimations of short-term or seasonal variations in the modelled density and the limitations of RF in extrapolating values outside the training data. Overall, our study presents a potential method for estimating unknown Antarctic firn densities using known densities and satellite parameters.

The purpose of this study is to evaluate the effectiveness of various machine learning algorithms in predicting the ultimate tensile strength (UTS) of friction stir welded joints. This prediction is crucial for assessing weld quality and integrity. Several investigations are carried out in linear and non-linear regression models, including Poisson Regressor, Gradient Boosting Regressor, Bayesian Ridge, k-Nearest Neighbours, Lasso, Random Forest, Elastic-Net, and Support Vector Regression, using datasets of welding parameters. The models were evaluated for their prediction accuracy and reliability using mean absolute error (MAE), mean square error (MSE) and R2 score metrics. The findings revealed significant variations in model performance. Non-linear models, especially Decision Trees and k-Nearest Neighbours, have exhibited the highest efficiency. The Decision Tree model achieved a remarkable R2 score of 0.97. Meanwhile, the k-Nearest Neighbours model noted the lowest MSE value of 408.62, whereas the MAE value was 14.75. In comparison, the Support Vector Regression model lagged significantly, registering the best MSE and MAE of 14824.74 and 107.73, along the side of a negative R2 score of −0.22, indicating deficient performance. This study highlights the potential of machine learning to forecast the quality of the weld joints. It proposes that the utilisation of these models can significantly enhance production processes and optimise material performance in industrial applications. The study demonstrates the integration of machine learning into manufacturing workflows to enable predictive maintenance and enhance quality control.

• ML models for predicting SCM concretes compressive strength developed • Multiple datasets devised from a global dataset were used for training • Multiple ML models used in conjunction with multiple datasets for best results • Accurate prediction results obtained for all SCM concrete types studied • Prediction uncertainty bounds established by conformal prediction analysis Supplementary cementitious materials (SCM) used as partial replacements of cement can reduce carbon emissions and construction costs. The amount and types of SCM used directly influence the compressive strength of concrete, which is a crucial indicator of concrete performance. However, predicting compressive strength is challenging for SCM-based concretes due to their complex compositions, especially when multiple types of SCM are used. This study used five machine learning (ML) models, namely artificial neural network (ANN), random forest (RF), extra trees (ET), extreme gradient boosting (XGB), and adaptive boosting (AdaBoost) to predict the 28-day compressive strength of SCM concretes incorporating fly ash (FA), silica fume (SF), and ground granulated blast furnace slag (GGBFS). Firstly, a global dataset comprising 1456 data points was organized into seven individual and six combined datasets for training of the five ML models. After this, the performances of the ML models’ predictions using individual and combined datasets were compared using root mean squared error (RMSE) and coefficient of determination (R 2 ) to identify the optimal ML model-dataset configurations for each type of SCM-based concrete. It is found that in general, XGB models developed using the combined datasets demonstrated superior prediction accuracy, achieving an R² value exceeding 0.8. Both Shapley Additive Explanation (SHAP) and partial dependence plots (PDPs) were employed to illustrate the effects of input variables on compressive strength. Finally, uncertainty quantification using the Jackknife method combined with a normalized nonconformity measure was employed to provide accurate uncertainty bounds for compressive strength predictions.

Accurate forecasting of photovoltaic (PV) generated electricity is essential for efficiently managing and integrating Renewable Energy (RE) into electricity distribution systems. This research investigation optimizes Feature Selection (FS) and prediction results for PV energy prediction by applying Bayesian Density Estimation (BDE) with Elastic Net (ELNET) regression analysis. This phenomenon and unacceptable outcomes are prevalent when applying conventional regression algorithms on datasets with significant results and addressing predictor multicollinearity. Improved FS and multicollinearity control has been rendered feasible by ELNET, which integrates the best features of Ridge and Lasso regression. ELNET eliminates these challenges through the implementation of L1 and L2 penalties. Non-parametric prediction Bayesian Density Estimation (BDE) is comprehensive data regarding residual distributions and predictor impacts. By incorporating ELNET’s regularisation and FS abilities with BDE’s statistical prediction and adaptability, the recommended ELNET-BDE is proposed to attain more accurate and reliable predictions. This technique has been used to assess massive data sets developing from Visakhapatnam, India, incorporating historical PV energy generation combined with definite Meteorological Factors (MF). Considering comprehensive data preliminary processing, FS, and validation, ELNET-BDE outperforms existing methods. Research investigations demonstrate that the ELNET-BDE model attains significantly lower Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) than contesting Machine Learning (ML) algorithms like Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting Machines (GBM). Compared with distinct FS techniques, RMSE can be minimized by up to 15% and MAE by up to 20%. The findings specify a substantial improvement in accuracy in prediction, emphasizing how the model can be used for improving solar power grid integration and energy for improved RE management.

Background/Objectives: Diabetic kidney disease (DKD) is a major cause of end-stage kidney disease and a leading contributor to morbidity and mortality in patients with type 2 diabetes mellitus (T2DM). However, predictive models for DKD onset in Korean patients with T2DM remain underexplored. This study aimed to develop and validate a machine learning (ML)-based DKD prediction model for this population. Methods: This retrospective study utilized electronic health records from six secondary or tertiary hospitals in Korea. The Jeonbuk National University Hospital cohort was used for model development (ratio training: test data, 8:2), whereas datasets from five other hospitals supported external validation. We employed multiple ML algorithms, including lasso, ridge, and elastic net regression; random forest; XGBoost; support vector machines; and neural networks. The model incorporated demographic variables, comorbidities, medications, and laboratory test results. Results: Among 5120 patients with T2DM, 1361 (26.6%) developed DKD. In the development cohort, XGBoost achieved the highest predictive performance (AUC: 0.8099), followed by random forest and logistic regression models (AUCs: 0.7977-0.8019). External validation confirmed the model's robustness with high AUCs (XGBoost: 0.8113, logistic regression models: 0.8228-0.8271). Key predictive factors included age; baseline estimated glomerular filtration rate; and creatinine, hemoglobin, and hemoglobin A1c levels. Conclusions: Our findings highlight the potential of ML-based approaches in predicting DKD in patients with T2DM. The superior performance of XGBoost and logistic regression models underscores their clinical utility. External validation supports the model's generalizability. This model is a valuable tool for the early DKD risk assessment of Korean patients with T2DM.

IntroductionThere has been an increased demand for more accurate prediction tools to aid clinical decision-making regarding disease diagnosis prognosis for coronary artery disease(CAD) patients. Patients undergoing CABG surgery are older and a larger number have had previous heart surgery. Consequently, mortality after CABG is expected to increase despite procedural advances.
 Objectives and ApproachThis study aims to compare the predictive performance of random forest(RF) and logistic regression(LR) classifiers for predicting 30-day and 1-year post-operative mortality risk in CAD patients who underwent CABG. Data was obtained by linking the Alberta Provincial Project for Outcome Assessment in Coronary Heart Disease(APPROACH) registry, a prospective longitudinal data of patients undergoing cardiac catheterization in Alberta, Canada, to vital statistics database. All patients who underwent first-time isolated CABG between January 1, 2007 and December 31, 2012 were included in the analysis. Area under the receiver operating curve(AUC) was used to compare the predictive performance of LR and RF regression.
 ResultsOf the 4,908 eligible subjects who underwent isolated CABG during the study period, mortality estimates of 30-day and 1-year post CABG surgery were 1.59% and 3.85%, respectively. Descriptive analysis revealed that age, sex, hypertension, dialysis, cerebrovascular disease, chronic obstructive pulmonary disease, and chronic heart failure were associated with 30-day and 1-year mortality. The accuracy of the LR and RF regression classifiers in predicting 30-day mortality were 74.1, and 99.7%, respectively. While the accuracy of the former and latter classifiers in predicting 1-year post CABG mortality were 74% and 97.4%, respectively.
 Conclusion/ImplicationsThis study shows that RF classifier results in better predictive accuracy than LR in predicting post-operating mortality risk in CAD patients. Machine learning models are potentially usefully for developing clinical prediction models that can be used to aid the monitoring of post-discharge outcomes in the management of cardiovascular diseases.

BackgroundEarly identification of dementia is crucial for prompt intervention for high-risk individuals in the general population. External validation studies on prognostic models for dementia have highlighted the need for updated models. The use of machine learning in dementia prediction is in its infancy and may improve predictive performance. The current study aimed to explore the difference in performance of machine learning algorithms compared to traditional statistical techniques, such as logistic and Cox regression, for prediction of all-cause dementia. Our secondary aim was to assess the feasibility of only using clinically accessible predictors rather than MRI predictors.MethodsData are from 4,793 participants in the population-based AGES-Reykjavik Study without dementia or mild cognitive impairment at baseline (mean age: 76 years, % female: 59%). Cognitive, biometric, and MRI assessments (total: 59 variables) were collected at baseline, with follow-up of incident dementia diagnoses for a maximum of 12 years. Machine learning algorithms included elastic net regression, random forest, support vector machine, and elastic net Cox regression. Traditional statistical methods for comparison were logistic and Cox regression. Model 1 was fit using all variables and model 2 was after feature selection using the Boruta package. A third model explored performance when leaving out neuroimaging markers (clinically accessible model). Ten-fold cross-validation, repeated ten times, was implemented during training. Upsampling was used to account for imbalanced data. Tuning parameters were optimized for recalibration automatically using the caret package in R.Results19% of participants developed all-cause dementia. Machine learning algorithms were comparable in performance to logistic regression in all three models. However, a slight added performance was observed in the elastic net Cox regression in the third model (c = 0.78, 95% CI: 0.78–0.78) compared to the traditional Cox regression (c = 0.75, 95% CI: 0.74–0.77).ConclusionsSupervised machine learning only showed added benefit when using survival techniques. Removing MRI markers did not significantly worsen our model’s performance. Further, we presented the use of a nomogram using machine learning methods, showing transportability for the use of machine learning models in clinical practice. External validation is needed to assess the use of this model in other populations. Identifying high-risk individuals will amplify prevention efforts and selection for clinical trials.

In cement manufacturing, ensuring consistent product quality remains a challenge due to variations in raw materials, operational conditions, and delays in laboratory testing, particularly compressive strength tests, which are only available after 3, 7, and 28 days. This study aims to address this issue by developing a predictive analytics model that estimates compressive strength using machine learning, based on early-available laboratory parameters. The research is conducted at PT Semen Pertama and uses the CRISP-DM framework to structure the analytical process, from business understanding to model deployment. Historical laboratory data—comprising chemical compositions (e.g., SiO₂, Al₂O₃, Fe₂O₃, CaO), physical properties (e.g., fineness, residue), and strength test results—were used to train two supervised learning models: Linear Regression and Random Forest Regressor. Several feature selection methods were applied to improve model accuracy and interpretability. Model performance was assessed using standard regression metrics and validated with cross-validation. The results show that Random Forest consistently achieved higher predictive accuracy than Linear Regression. Feature importance analysis highlighted key variables influencing compressive strength, providing practical insights for quality monitoring. This study supports earlier quality estimation and proactive decision-making in production.

The aim of this research is to predict preplaced-aggregate concrete (PAC) compressive strength (CS) by using machine learning approaches such as gene expression programming (GEP) and random forest (RF). PAC requires injecting a portland cement-sand grout with admixtures into a mold after coarse aggregate has been deposited, making CS prediction complicated and requiring substantial study. Machine learning methods were used to cut down on the time and money needed for extensive experimental testing. The database includes 135 values for CS with eleven input variables. There is an acceptable degree of agreement between predicted and experimental values, as shown by the CS R2 values of 0.94 for GEP and 0.96 for RF. When comparing RF with GEP, RF performed better as measured by R2. The lower values displayed by the statistical error also showed that RF performed better than GEP. To compare, the GEP model's COV, MAE, RSME, and RMSLE were 0.527, 1.569, 2.706, and 0.133, whereas those for RF were 0.450, 1.648, 2.17, and 0.092. The SHAP analysis showed the effects of each input parameter, illuminating the positive effect of increasing the superplasticizer content on strength and the negative effect of raising the water-to-binder ratio. Using machine learning approaches to forecast the CS of PAC, this study has the potential to boost environmental protection and economic advantage.

Background and Objectives: Hospital length of stay (LOS) after breast cancer surgery is a key indicator of postoperative recovery, healthcare quality, and hospital resource utilization. Traditional statistical approaches have identified general correlates of LOS but remain limited in predictive accuracy, particularly in heterogeneous real-world surgical populations. Machine learning (ML) models may offer improved performance by capturing nonlinear interactions among clinical, pathological, and operative factors. This study aimed to evaluate ML algorithms for LOS prediction and to identify determinants of prolonged hospitalization in a contemporary breast cancer cohort. Materials and Methods: We conducted a retrospective cross-sectional study of 198 consecutive breast cancer patients who underwent surgery between January 2022 and December 2023 at a single tertiary care center. Clinical, pathological, and surgical data were extracted from electronic medical records. Three regression models—multiple linear regression, Random Forest, and Gradient Boosting—were trained to predict continuous LOS, and three classification models were applied to prolonged LOS (≥10 days). Model performance was assessed using mean absolute error (MAE), root mean square error (RMSE), coefficient of determination (R2), and area under the curve (AUC). Feature importance was analyzed for the best-performing model. Results: The median LOS was 7 days (IQR 5–10), ranging from 1 to 26 days. Breast-conserving surgery showed the shortest LOS (median 3 days), while mastectomy with immediate reconstruction resulted in the longest stays (median 8 days). Random Forest regression achieved the lowest prediction error (MAE 2.31 days; RMSE 2.82; R2 = 0.37), outperforming Gradient Boosting and substantially surpassing linear regression (MAE 8.63 days; R2 = –8.17). Key predictors included age, surgical complexity, reconstruction modality, BMI, implant capacity, and tumor burden. Classification models yielded modest AUCs (0.545–0.589) with low sensitivity, indicating limited discriminative performance for dichotomized LOS outcomes. Conclusions: Machine-learning models, particularly Random Forest, substantially improve LOS prediction compared with classical regression and provide clinically meaningful insights into the drivers of hospitalization after breast cancer surgery. Continuous LOS modeling is more informative than binary thresholds. These findings support integrating ML-based tools into perioperative planning, resource allocation, and patient counseling in breast surgical care.

Malawi’s Miombo Woodlands provide critical wood products and ecosystem services, including carbon storage through above ground-biomass (AGB). However, accurate and efficient estimation of AGB and C remains a huge challenge due to the complexity of heterogeneous natural and anthropogenic environmental factors exacerbated by choice of data collection and analysis techniques. A study was conducted to explore the potential of Machine Learning (ML) algorithmic on modelling AGB in Malawi’s Miombo Woodlands using remotely sensed data. A combination of AGB field measurements and Sentinel 2 Multi-Spectral Instrument (S2 MSI) imagery, and environmental data were employed to train and evaluate the performance of Random Forest (RF) regression ML model, thus against the backdrop of traditional simple linear and multiple linear regression models used in the past. The randomForest package in Rstudio R-3.6.1 software was used to model the S2 median seasonal composite band imagery datasets acquired in 2019 over Ntchisi Forest. Results demonstrate an outstanding performance of the RF with Mean Decrease in Accuracy (MDA) importance of >80 acquired from the Green, Blue and Near-Infra-Red (NIR) bands of the post-rainy season imagery while the dry season scene registered a MDA importance of >52 obtained from the Red, Blue and NIR bands. These findings underscore the superiority of the RF, revealing its ability to model AGB using different spectral bands, and more importantly, in different seasons. The outcomes have also revealed the better accuracy that the RF render in generating AGB maps (providing a minimum range of 25-33 tCha-1, which is in line with the ground reference estimates). Overall, the study has shown that RF Regression is a better technique to estimate AGB in Miombo Woodlands using the S2 MSI imagery under the given conditions. In conclusion, the research has further shown that the optical band combination of NIR and the Red regions of the Electro-Magnetic Spectrum are indispensable variables for successful modelling of AGB as well as delineating forest cover from non-forest attributes in the Miombo Woodlands. However, the study recommends exploring modelling the AGB using averaged vegetation indices i.e., NDVI, SAVI, and MSAVI, for robust results.