Data-driven Prediction Research Articles

Data-driven prediction of electrospun nanofiber diameter using machine learning: A comprehensive study and web-based tool development

Electrospinning has emerged as a versatile technique for fabricating polymer nanofibers with diameters ranging from tens to hundreds of nanometers. Precisely controlling and predicting the nanofiber diameter is crucial for various applications. The complex interplay of multiple electrospinning parameters presents a significant challenge in this endeavor. This study introduces a novel data-driven approach using machine learning (ML) to predict and optimize the diameter of electrospun nanofibers. A comprehensive dataset of approximately 430 data points was compiled from literature sources, including polymer properties and electrospinning process parameters. Six ML algorithms were evaluated: Decision Tree (DT), Extra Trees Regression (ETR), Kernel Ridge Regression (KRR), Random Forest (RF), Support Vector Regression (SVR), and eXtreme Gradient Boosting (XGB). A 75/25 train-test split was selected for optimal model evaluation, with 10-fold cross-validation employed for hyperparameter tuning and estimating the variability of performance metrics. RF and ETR demonstrated the best predictive accuracy, with R2 values of 0.9468 and 0.9421, and RMSE values of 92.3 nm and 96.1 nm, respectively, on the test set. Feature importance analysis and SHapley Additive exPlanations (SHAP) values revealed that polymer concentration, applied voltage, and feed rate significantly influence the nanofiber diameter, which was further elucidated by partial dependence plots. To facilitate accessibility and collaboration, an interactive web server called ENDP was developed, allowing users to input polymer properties and process parameters to predict the nanofiber diameter. This study showcases the potential of ML in guiding the design and optimization of electrospun nanofibers, providing valuable insights for tailoring their properties into specific applications.

Data-driven prediction model for pollution prevention deficiencies on ships

Preventing ship-related environmental pollution is one of the most important concerns of the maritime industry. This research developed a machine learning-based model to detect pollution prevention deficiencies in detained ships to address critical environmental regulations and enhance maritime safety. To develop predictive model, the Random Forest algorithm was performed on the port state control report of 4056 detained ships in the Tokyo MoU region between 2019 and 2023. Quantitative analysis revealed that the port state conducting the inspection was the most impactful variable on the model, affecting its accuracy significantly due to diverse enforcement standards. Deficiency number was the second important variable. The analysis also revealed that ship type and size had above-average effect on the model. The Random Forest model achieved the highest accuracy of 72.43 % compared to other algorithms. Following the predictive modeling, Association Rule Mining was utilized to uncover and delineate implicit relationships among the identified deficiency areas, thus deepening our analysis of pollution prevention strategies. The Random Forest algorithm has high accuracy and robustness in models developed for big data and is resistant to overfitting, and the Apriori algorithm can extract hidden and complex relationships in the data faster and more easily interpretable were the main reasons for the selection of these algorithms. The Apriori algorithm analysis identified critical deficiency areas linked to pollution prevention—namely fire safety, certificate/documentation, ISM, emergency systems, and life-saving appliances—with these areas appearing in most of the rules with high metric values, thus highlighting common compliance gaps. The findings can help port authorities and port state control officers prioritize inspections of ships most likely to be non-compliant, thereby effectively reducing pollution prevention risk and inspection times and costs. Therefore, the measures to be taken may prevent ship-related pollution and achieve the United Nations Sustainable Development Goals, especially related to marine resources.

Data-driven prediction of wind pressure on low-rise buildings in complex heterogeneous terrains

This study presents a data-driven methodology for predicting the pressure coefficient statistics on the windward wall, roof, and leeward wall of low-rise buildings situated downwind of complex heterogeneous terrains. Two types of artificial neural network models were developed: the empirical parameter-based ANN (PANN) and the morphology-based ANN (MANN). Pressure data from wind tunnel tests on the Wind Engineering Research Field Laboratory (WERFL) building model (building height H = 4 m) in complex heterogeneous terrain were used to develop the ANN models. These models were evaluated against a non-linear fitted model to assess their predictive performances. PANN and MANN demonstrated superior performance in capturing the effects of terrain complexity on the mean (Cp,mean) and the root-mean-square (Cp,RMS) wind pressure coefficients for the windward wall, roof, and leeward wall. Optimal prediction was achieved with a terrain patch size of W × L = 4 × 2, equating to a full-scale area of approximately 72 m × 23 m. This suggests that the morphology within approximately 100 m × 50 m (25H× 12.5H) in front of a low-rise building has the greatest correlation with the wind pressure coefficient. Despite lower R2 values for max Cp,RMS on the leeward wall across all models, both PANN and MANN showed promising accuracy for the six outputs studied. Moreover, a global sensitivity analysis confirmed the impact of terrain roughness and complexity on the prediction models particularly on max Cp,RMS, and underscored the dominance of effective roughness length z0,eff and the coefficient of variation of roughness length COVz0 in influencing model outcomes.

Evaporating temperature estimation of refrigeration systems based on vibration data-driven soft sensors

The evaluation of the operating conditions of refrigeration compressors once installed in household appliances is challenging due to the need to install pressure transducers, a process which requires system evacuation and refrigerant reintroduction. In addition, changes in the piping modify the characteristics of the original product. This paper proposes a soft-sensing technique based on vibration measurements of the compressor surface to predict the evaporating temperature. Different machine learning (ML) techniques are evaluated as data-driven prediction models, namely multilayer perceptron (MLP) neural networks, least squares boosting, generalized additive model, random forest, extreme learning machine, and random vector functional link neural networks. These techniques were applied to data obtained from a test rig designed to emulate compressor operation in a refrigeration system, with an operating envelope from -30°C to -10°C for the evaporating temperature and from 34°C to 54°C for the condensing temperature. The results showed that, with a single vibration measurement point, it was possible to use an MLP technique to estimate the evaporating temperature with a root mean squared error of 1.74°C in a non-intrusive way. For the other prediction techniques, the errors were a bit higher than for the MLP, but the maximum error value was about 2.5°C in all cases.

Diagnosis of acute hyperglycemia based on data-driven prediction models

Acute hyperglycemia is a common endocrine and metabolic disorder that seriously threatens the health and life of patients. Exploring effective diagnostic methods and treatment strategies for acute hyperglycemia to improve treatment quality and patient satisfaction is currently one of the hotspots and difficulties in medical research. This article introduced a method for diagnosing acute hyperglycemia based on data-driven prediction models. In the experiment, clinical data from 1000 patients with acute hyperglycemia were collected. Through data cleaning and feature engineering, 10 features related to acute hyperglycemia were selected, including BMI (Body Mass Index), TG (triacylglycerol), HDL-C (High-density lipoprotein cholesterol), etc. The support vector machine (SVM) model was used for training and testing. The experimental results showed that the SVM model can effectively predict the occurrence of acute hyperglycemia, with an average accuracy of 96 %, a recall rate of 84 %, and an F1 value of 89 %. The diagnostic method for acute hyperglycemia based on data-driven prediction models has a certain reference value, which can be used as a clinical auxiliary diagnostic tool to improve the early diagnosis and treatment success rate of acute hyperglycemia patients.

Data-driven prediction of convective heat transfer coefficients in internal walls of aero-engine bearing chambers using Mind Evolution Algorithm-Enhanced Bayesian regularization neural networks

In the bearing chambers of aero-engines, the complex nature of the oil–gas flow complicates the precise calculation of heat transfer coefficients. Addressing computational discrepancies, firstly, this study establishes a correlation analysis method, which considers the relationship between several independent variables including the geometric shape of the bearing chamber, rotational speed, oil flow rate, thermal sealing air mass, circumferential angle, axial position, and the dependent variable, which is the logarithmic Nusselt number on the internal wall. Secondly, four models are developed using Multiple Linear Regression, Back Propagation Neural Network, Deep Belief Neural Network, and Convolutional Neural Network with Long Short-Term Memory Network, yielding Mean Relative Percentage Errors (MRPE) of 5.14%, 2.80%, 4.66%, and 4.12%, respectively. Due to the limited predictive performance of these models, the neural network topology is further optimized. After repeated trial-and-error calculations, the neural network with three hidden layers was found to be the most effective. Finally, a novel BP Neural Network optimized by the Mind Evolution Algorithm-enhanced with Bayesian regularization (MEA-BPNN-Bayesian) model is proposed forpredicting the local convective heat transfer, achieving an MRPE of 1.71%. The model accurately predicted heat transfer coefficients, opening up new ways to improve heat management in bearing chambers.

Surrogate gradient methods for data-driven foundry energy consumption optimization

In many industrial applications, data-driven models are more and more commonly employed as an alternative to classical analytical descriptions or simulations. In particular, such models are often used to predict the outcome of an industrial process with respect to specific quality characteristics from both observed process parameters and control variables. A major step in proceeding from purely predictive to prescriptive analytics, i.e., towards leveraging data-driven models for process optimization, consists of, for given process parameters, determining control variable values such that the output quality improves according to the process model. This task naturally leads to a constrained optimization problem for data-driven prediction algorithms. In many cases, however, the best available models suffer from a lack of regularity: methods such as gradient boosting or random forests are generally non-differentiable and might even exhibit discontinuities. The optimization of these models would therefore require the use of derivative-free techniques. Here, we discuss the use of alternative, independently trained differentiable machine learning models as a surrogate during the optimization procedure. While these alternatives are generally less accurate representations of the actual process, the possibility of employing derivative-based optimization methods provides major advantages in terms of computational performance. Using classical benchmarks as well as a real-world dataset obtained from an industrial environment, we demonstrate that these advantages can outweigh the additional model error, especially in real-time applications.

Spectral prediction method based on the transformer neural network for high-fidelity color reproduction

Color distortion often occurs during transmission and reproduction processes, and existing spectral prediction methods have the disadvantage of low prediction accuracy in halftone reproduction. Addressing this issue, this paper establishes a halftone dataset composed of four-color inks (CMYK) mixtures. Based on this, the transformer network is introduced to model and characterize the spectral features of mixed inks, and a forward color formulation prediction model and a reverse spectral prediction model combining halftone reproduction with spectral sequences are proposed, namely the spectrum-color transformer (SC-Former). Color reproduction quality assessment experiments are conducted using the dataset established in this paper and the international standard Ugra/Fogra Media Wedge V3.0 test set. The experimental results show that the SC-Former model outperforms traditional physical models and data-driven prediction models in terms of color reproduction effects and spectral prediction accuracy. This research contributes to the development of high-fidelity color reproduction techniques.

Investigation of Machine Learning Regression Techniques to Predict Critical Heat Flux Over a Large Parameter Space

A unifying and accurate model to predict critical heat flux (CHF) over a wide range of conditions has been elusive since wall boiling research emerged. With the release of the experimental data utilized in the development of the 2006 Groeneveld CHF lookup table (LUT), by far the most extensive public CHF database available to date (nearly 25 000 data points), the development of data-driven prediction models over a large parameter space in simple geometry (vertical, uniformly heated round tubes) can be achieved. Furthermore, the popularization of machine learning techniques to solve regression problems has led to more advanced tools for analyzing large and complex databases. This work compares three machine learning algorithms to predict the CHF database. For each selected regression algorithm (ν-Support vector, Gaussian process, and neural network), a set of optimized hyperparameters is applied. Among the investigated algorithms, the neural network emerges as the most effective, achieving a CHF predicted/measured factor standard deviation of 12.3%, three times less than that of the LUT. In comparison, the Gaussian process regression and the ν-support vector regression achieve a standard deviation of 17.7%, about two times lower than the LUT. Hence, all considered algorithms significantly outperform the LUT prediction performance. The neural network model and training methodology are designed to prevent overfitting, which is confirmed by data analyses of the CHF predictions. Finally, future development directions (including data coverage, transfer learning, and uncertainty quantification) are discussed.

Prediction of segmental motor outcomes in traumatic spinal cord injury: Advances beyond sum scores

Background and objectivesNeurological and functional recovery after traumatic spinal cord injury (SCI) is highly challenged by the level of the lesion and the high heterogeneity in severity (different degrees of in/complete SCI) and spinal cord syndromes (hemi-, ant-, central-, and posterior cord). So far outcome predictions in clinical trials are limited in targeting sum motor scores of the upper (UEMS) and lower limb (LEMS) while neglecting that the distribution of motor function is essential for functional outcomes. The development of data-driven prediction models of detailed segmental motor recovery for all spinal segments from the level of lesion towards the lowest motor segments will improve the design of rehabilitation programs and the sensitivity of clinical trials. MethodsThis study used acute-phase International Standards for Neurological Classification of SCI exams to forecast 6-month recovery of segmental motor scores as the primary evaluation endpoint. Secondary endpoints included severity grade improvement, independent walking, and self-care ability. Different similarity metrics were explored for k-nearest neighbor (kNN) matching within 1267 patients from the European Multicenter Study about Spinal Cord Injury before validation in 411 patients from the Sygen trial. The kNN performance was compared to linear and logistic regression models. ResultsWe obtained a population-wide root-mean-squared error (RMSE) in motor score sequence of 0.76(0.14, 2.77) and competitive functional score predictions (AUCwalker = 0.92, AUCself-carer = 0.83) for the kNN algorithm, improving beyond the linear regression task (RMSElinear = 0.98(0.22, 2.57)). The validation cohort showed comparable results (RMSE = 0.75(0.13, 2.57), AUCwalker = 0.92). We deploy the final historic control model as a web tool for easy user interaction (https://hicsci.ethz.ch/). DiscussionOur approach is the first to provide predictions across all motor segments independent of the level and severity of SCI. We provide a machine learning concept that is highly interpretable, i.e. the prediction formation process is transparent, that has been validated across European and American data sets, and provides reliable and validated algorithms to incorporate external control data to increase sensitivity and feasibility of multinational clinical trials.

Data-driven prediction of spinal cord injury recovery: An exploration of current status and future perspectives

Spinal Cord Injury (SCI) presents a significant challenge in rehabilitation medicine, with recovery outcomes varying widely among individuals. Machine learning (ML) is a promising approach to enhance the prediction of recovery trajectories, but its integration into clinical practice requires a thorough understanding of its efficacy and applicability. We systematically reviewed the current literature on data-driven models of SCI recovery prediction. The included studies were evaluated based on a range of criteria assessing the approach, implementation, input data preferences, and the clinical outcomes aimed to forecast. We observe a tendency to utilize routinely acquired data, such as International Standards for Neurological Classification of SCI (ISNCSCI), imaging, and demographics, for the prediction of functional outcomes derived from the Spinal Cord Independence Measure (SCIM) III and Functional Independence Measure (FIM) scores with a focus on motor ability. Although there has been an increasing interest in data-driven studies over time, traditional machine learning architectures, such as linear regression and tree-based approaches, remained the overwhelmingly popular choices for implementation. This implies ample opportunities for exploring architectures addressing the challenges of predicting SCI recovery, including techniques for learning from limited longitudinal data, improving generalizability, and enhancing reproducibility. We conclude with a perspective, highlighting possible future directions for data-driven SCI recovery prediction and drawing parallels to other application fields in terms of diverse data types (imaging, tabular, sequential, multimodal), data challenges (limited, missing, longitudinal data), and algorithmic needs (causal inference, robustness).

Data-driven prediction of future melt pool from built parts during metal additive manufacturing

Smart manufacturing in metal additive manufacturing (MAM) relies on real-time process optimization through the prediction of unbuilt parts using data from built parts. Melt pool dynamics are intimately associated with the development of various microstructures during the MAM. In this paper, we demonstrate the idea by establishing a machine learning (ML) model to predict the melt pool of future parts, based on the experimental data from the National Institute of Standards and Technology (NIST) where 80 % is used for the training (built parts) and 20 % for testing (future parts). The ML model integrates both data denoising and predictive modeling. First, a convolutional neural network (CNN) is used to process a large dataset of raw melt pool images, enabling the automatic removal of noises (e.g., splash and plume) and thereof extraction of high-quality melt pool data. Following that, a novel data-driven melt pool model based on multi-layer perceptron (MLP) is trained by incorporating raw, long scanning history as input features, which best accounts for the effects of printing history (e.g., intertrack heating) on melt pool development. It takes complete advantage of MLP in handling high-dimensional regression problems in conjunction with a large dataset. Upon testing under various manufacturing conditions, the average relative error magnitude (AREM) of predicting melt pool size drops to 2.8 %, compared to 14.8 % of the prior art – the Neighboring Effect Modeling Method model (NBEM). This research thus represents a significant step towards reliable melt-pool-guided AM process optimization for smart manufacturing, enabled by advanced, flexible, and maximum use of ML techniques.

Data-driven prediction of rail neutral temperature for continuously welded rails using impulse-based vibration frequencies

Continuously welded rails (CWR) are prone to the development of high thermal-induced load along the axial direction. Excessive levels of load lead to risk of rail buckling and potential for derailment. Knowledge of the in situ rail axial load in CWRs is therefore important to ensure safe rail management. Field-deployable, nondestructive evaluation techniques for measuring the rail load, or a widely adopted alternative called rail neutral temperature (RNT), are desired. This study uses a data-driven approach to investigate if rail dynamic response data, collected in a non-destructive fashion, can be used to predict RNT. The study is based on a data set comprising rail equivalent strain, temperature and vibration resonance frequencies that was collected from a revenue-service rail over a period of nearly two years. All excited vibration resonance peaks are identified from other peaks caused by noise using spectral amplitude variance. Among these resonance peaks, potentially useful resonances are identified with respect to stacked spectra collected across a testing day using an assumed temperature-frequency relation. A subset of the identified useful resonances is then identified based on their consistent appearance across both testing locations and all testing days, strong correlation to effective strain, and strong correlation to each other. Three particular vibration resonances (or vibration modes -- these terms will be used interchangeably throughout this paper unless specified otherwise. The term mode does not necessarily indicate mode shapes or mode families.) emerge from this process as best candidates. A classic feature selection technique, Lasso linear regression, is then employed to identify critical power combinations of the three resonant mode frequencies. Two power combinations exhibit unique correlation to the measured equivalent axial strain at both test locations across all testing days, and thus show particular ability to predict RNT. The RNT is predicted at one test location using different models based on the power combination data from the other location, and vice versa, where the predictions satisfy standard RNT measurement accuracy expectations.

Data-driven prediction of strain fields in auxetic structures and non-contact validation with mechanoluminescence for structural health monitoring

Recent advancements in 3D printing technology have significantly enhanced the potential of auxetic structures, which are notable for their negative Poisson’s ratio, particularly in applications such as sensor technology and structural health monitoring. Central to the performance of these structures is the accurate estimation of the effective strain parameter, a critical metric for assessing structural integrity. However, as structural complexity increases, estimating this parameter becomes increasingly challenging. The fabrication and real-world validation of these structures are equally important challenges. This paper introduces two key innovations for the practical application of auxetic structures. First, we present a multi-kernel hierarchical deep neural network that leverages finite element simulation data to accurately predict effective strain fields in complex auxetic configurations. This model architecture not only reduces the number of parameters requiring training but also enhances feature learning and generalization capabilities, achieving over 90% accuracy in predicting strain fields. Second, we validate these predictions using a 3D-printed specimen embedded with mechanoluminescent (ML) particles. This approach enables direct, non-contact visualization of strain in real-time, offering high spatial and temporal resolution. The alignment observed between predicted and observed strain concentration areas demonstrates the efficacy of integrating ML technology into auxetic designs. This integration significantly improves the reliability and diagnostic capabilities of advanced structural systems.

Experimental and numerical study on data-driven prediction for wildfire spread incorporating adaptive observation error adjustment

In recent wildfire prediction research, data assimilation (DA) methods like Ensemble Kalman filtering have gained traction for integrating observation data to enhance prediction accuracy. Most previous studies trusted that the observation data were accurate, and set a small observation error, which causes unreliable predicted results for scenarios with large observation error. To tackle this, our study introduced a method that iteratively adjusted the potential range of observation errors by comparing observation and simulation data over time. We conducted a 30-m experiment and kilometer-scale numerical simulations. Unlike prior research, we adopted larger error ranges (the similarity index with true data ranges from 0.6 to 1) for both real and synthetic observation data. In the experiment, to increase the complexity of fire spread, a heterogeneous fuel arrangement was employed. Irregular flame fronts appeared due to incomplete combustion and were difficult to replicate in simulations. Better accuracy was achieved using real observation data to revise predictions. Furthermore, to improve the applicability of the algorithm, numerical simulations were designed to consider observation error changing over time or not. The Root Mean Square Errors for the fire front prediction using the proposed method remained lower than that of the traditional DA approach.

Investigation of a solar-assisted methanol steam reforming system: Operational factor screening and computational fluid dynamics data-driven prediction

The development of solar-to-fuel technologies has received significant attention while facing the growing interest in clean energy production. To accurately evaluate and optimize the solar-assisted thermochemical hydrogen production process, a reliable prediction method that ensures stable accuracy is essential. We utilized a gray correlation analysis (GRA) and a multi-objective genetic algorithm (GA) model to optimize the operating parameters of a methanol steam reforming system. Computational fluid dynamics simulations provided the response values for key parameters, including reactant conversion rate, syngas product yield, and CO selectivity. Additionally, the Taguchi’s method and analysis of variance (ANOVA) were employed to assess the impact of operating parameters on these target outputs. The results demonstrate that the proposed GA-Back propagation neural networks (GA-BPNN) model significantly outperforms traditional prediction models in accuracy, exhibiting robustness and generalization. The mean absolute percentage error (MAPE) of methanol conversion, hydrogen yield, and CO selectivity is 3.20 %, 4.69 % and 3.53 %. The GRA-GA-BPNN model shows a slight decline, while still maintains reliable prediction accuracy, with the MAPE value of 5.76 %, 5.11 %, and 10.4 %. This operational factor screening-based prediction model proves useful for large-scale data-driven predictions of solar thermochemical systems, especially under complex and variable external conditions and internal human interventions.

Data-driven prediction of tool wear using Bayesian regularized artificial neural networks

The prediction of wear in cutting tools is pivotal for boosting productivity and reducing manufacturing costs. Although current data-driven models in machine learning and deep learning have advanced predictive capabilities, they often lack generality and demand substantial data training. This paper presents a novel approach using Bayesian Regularized Artificial Neural Networks (BRANNs) to precisely forecast wear in milling tools. Unlike conventional machine learning models, BRANNs merge the strengths of artificial neural networks (ANNs) and Bayesian regularization, yielding a more robust and generalized predictive model. We utilized three open-access datasets from the literature alongside an in-house dataset generated by our milling setup. Initially, we assessed the model’s predictive ability by training and testing it against individual open-access datasets. We investigated the impact of input features, training data size, hidden units, training algorithms, and transfer functions on the model’s predictive capability. Subsequently, we trained the model using three open-access datasets and tested it against our in-house data. Our findings demonstrate that the developed model surpasses existing state-of-the-art models in accuracy and transferability

Bridging Machine Learning and Thermodynamics for Accurate pK a Prediction.

Integrating scientific principles into machine learning models to enhance their predictive performance and generalizability is a central challenge in the development of AI for Science. Herein, we introduce Uni-pK a, a novel framework that successfully incorporates thermodynamic principles into machine learning modeling, achieving high-precision predictions of acid dissociation constants (pK a), a crucial task in the rational design of drugs and catalysts, as well as a modeling challenge in computational physical chemistry for small organic molecules. Uni-pK a utilizes a comprehensive free energy model to represent molecular protonation equilibria accurately. It features a structure enumerator that reconstructs molecular configurations from pK a data, coupled with a neural network that functions as a free energy predictor, ensuring high-throughput, data-driven prediction while preserving thermodynamic consistency. Employing a pretraining-finetuning strategy with both predicted and experimental pK a data, Uni-pK a not only achieves state-of-the-art accuracy in chemoinformatics but also shows comparable precision to quantum mechanics-based methods.

Digital twinning, prediction and multi-objective optimization of an azeotrope system separation process in pharmaceutical manufacturing process

Digital twin is to make full use of the physical model, sensor update, operation history and other data to integrate multidisciplinary, multi-physical quantity, multi-scale and multi-probability simulation process, and to complete the full life cycle process of the corresponding physical equipment in the virtual space. In this paper, digital twin technology was used to simulate the production process of a pharmaceutical workshop, thus realizing solvent recovery by an azeotrope system separation process. The separation was optimized by multi-objective genetic algorithm. separation process, wherein the polynomial curve fitting and entropy-weighted TOPSIS data evaluation algorithms were combined to optimize the operating parameters, then the solvent purity and yield reach 99.96 % and 77.9 %, respectively, which is much higher than the corporate requirements. In addition, the effectiveness of the digital twin in providing data-driven prediction capability was verified by a CNN-LSTM-based regression prediction model, and the CNN-LSTM model has an accuracy of more than 98.9 % for all prediction results. Finally, a 3D visualization model was built using the Unity engine, enabling researchers to visualize and explore changes in key metrics during the separation process. This study can provide some guidance for digital transformation and intelligent development in pharmaceutical industry.

Data-driven prediction on critical mechanical properties of engineered cementitious composites based on machine learning

The present study introduces a novel approach utilizing machine learning techniques to predict the crucial mechanical properties of engineered cementitious composites (ECCs), spanning from typical to exceptionally high strength levels. These properties, including compressive strength, flexural strength, tensile strength, and tensile strain capacity, can not only be predicted but also precisely estimated. The investigation encompassed a meticulous compilation and examination of 1532 datasets sourced from pertinent research. Four machine learning algorithms, linear regression (LR), K nearest neighbors (KNN), random forest (RF), and extreme gradient boosting (XGB), were used to establish the prediction model of ECC mechanical properties and determine the optimal model. The optimal model was utilized to employ SHapley Additive exPlanations (SHAP) for scrutinizing feature importance and conducting an in-depth parametric analysis. Subsequently, a comprehensive control strategy was devised for ECC mechanical properties. This strategy can provide actionable guidance for ECC design, equipping engineers and professionals in civil engineering and material science to make informed decisions throughout their design endeavors. The results show that the RF model demonstrated the highest prediction accuracy for compressive strength and flexural strength, with R2 values of 0.92 and 0.91 on the test set. The XGB model outperformed in predicting tensile strength and tensile strain capacity, with R2 values of 0.87 and 0.80 on the test set, respectively. The prediction of tensile strain capacity was the least accurate. Meanwhile, the MAE of the tensile strain capacity was a mere 0.84%, smaller than the variability (1.77%) of the test results in previous research. Compressive strength and tensile strength demonstrated high sensitivity to variations in both water-cement ratio (W) and water reducer (WR). In contrast, flexural strength exhibited high sensitivity solely to changes in W. Conversely, the sensitivity of tensile strain capacity to input features was moderate and consistent. The mechanical attributes of ECC emerged from the combined effects of multiple positive and negative features. Notably, WR exerted the most significant influence on compressive strength among all features, whereas polyethylene (PE) fiber emerged as the primary driver affecting flexural strength, tensile strength, and tensile strain capacity.