Stack Model Research Articles

Loan Defaults Prediction Based on Stacked Models Trained by Personalized Features

Training one machine learning model with features that all clients have will result in a waste of features, which is likely to adversely affect the model’s performance. To solve the problem, the study attempts a new method, which it to train an individual stacked model for each loan client based on personalised features. Data used contains information of about fifteen million loan applicants, their default status, and 468 features in all. 41 of the features that can be quantitively analysed are selected according to the feature importance output by a Random Forest model. Default prediction of every client is made by a stacked model trained with all selected features he/she has. The stacked model consists of two layers, in which a Light Gradient-Boosting Machine (LGBM) classifier is the base learner, and a Logistic Regression model is the meta learner. As the defaulters account for only 3.14%, which is significantly unbalanced, Area Under the Curve (AUC) and F1 scores are employed to evaluate the method, instead of accuracy. Test results show that models trained by personalised features perform better than the ones trained by shared features. Additionally, the stacked model outperforms individual Logistic Regression model, but performs nearly the same as individual LGBM Classifier. In detailed, the stacked models trained with personalised features result in AUC=0.772 and F1=0.188. Due to data unbalance, although the method’s F1 score is relatively low, it’s considered to be passable. In the future, stacked models combining different models will be attempted.

A prediction model for heat exchanger fouling factor based on stacking model

A prediction model for heat exchanger fouling factor based on stacking model

RetNet30: A Novel Stacked Convolution Neural Network Model for Automated Retinal Disease Diagnosis

ABSTRACTAutomated diagnosis of retinal diseases holds significant promise in enhancing healthcare efficiency and patient outcomes. However, existing methods often lack the accuracy and efficiency required for timely disease detection. To address this gap, we introduce RetNet30, a novel stacked convolutional neural network (CNN) designed to revolutionize automated retinal disease diagnosis. RetNet30 combines a custom‐built 30‐layer CNN with a fine‐tuned Inception V3 model, integrating these sub‐models through logistic regression to achieve superior classification performance. Extensive evaluations on retinal image datasets such as DRIVE, STARE, CHASE_DB1, and HRF demonstrate significant improvements in accuracy, sensitivity, specificity, and area under the ROC curve (AUROC) when compared to conventional approaches. By leveraging advanced deep learning architectures, RetNet30 not only enhances diagnostic precision but also generalizes effectively across diverse datasets, establishing a new benchmark in retinal disease classification. This novel approach offers a highly efficient and reliable solution for early disease detection and patient management, addressing the limitations of manual examination methods. Through rigorous quantitative and qualitative assessments, our proposed method demonstrates its potential to significantly impact medical image analysis and improve healthcare outcomes. RetNet30 marks a major step forward in automated retinal disease diagnosis, showcasing the future of AI‐driven advancements in ophthalmology.

A stacked LSTM model for day-ahead solar irradiance forecasting under tropical seasons in Java-Bali

A stacked LSTM model for day-ahead solar irradiance forecasting under tropical seasons in Java-Bali

Predicting non-suicidal self-injury among Chinese adolescents: The application of ten algorithms of machine learning

Background and aimsHigh non-suicidal self-injury (NSSI) prevalence among adolescents is a global health issue. However, current prediction models for adolescent NSSI rely on a limited set of algorithms, resulting in biased predictions. Therefore, the aim of this study is to develop multiple machine learning models to enhance prediction accuracy and mitigate biases among Chinese adolescents. MethodsA total of 4487 junior and senior high school students in China were recruited. Multiple algorithms were included, such as logistic regression, decision tree, support vector machine, Naive Bayes, multi-layer perceptron, K-nearest neighbors, and ensemble learning algorithm like random forest, bagging, AdaBoost, and stacking to build predictive models. Data processing techniques, including standardization and the synthetic minority oversampling technique, were employed to optimize the predictive model. The model was trained on 70 % of the data, reserving 30 % for testing. ResultsThe ten prediction models achieved a good performance, with area under the receiver operating characteristic curve (AUC) scores above 0.700 in the test set. The stacking and random forest models achieved AUC scores of 0.904 and 0.898, respectively. The prediction performance of the Naive Bayes model was relatively poor. The top five important variables were resilience, bully, suicidal ideation, internet addiction, and depression. ConclusionsThe ensemble machine learning algorithm showed promising results predicting NSSI among adolescents. Such algorithms should be recommended for future NSSI research to enhance predictive accuracy. Identification of important features in NSSI prediction can help develop screening protocols and lay a foundation for clinical diagnosis and intervention in adolescent populations.

Gunshot Wounds to the Head: 29 Cases of Monocenter Children's Trauma Hospital Experience and Predictor Scales in Pediatric Population

Gunshot wounds to the head (GSWTH) in children remain an underexplored area, and all clinical guidelines extrapolated from adult experiences. A key challenge in treating these patients is age stratification, as pediatric survival rates are notably higher than in adults. The objective of the study is to compare 2 groups of patients based on the severity of their condition and to analyze the impact of various factors on the outcomes of these conditions. A single-center retrospective cohort study was conducted. Patient records of individuals (from birth to 18years old) with craniocerebral injuries caused by GSWTH and treated at Morozov Children's City Hospital between 1992 and 2015 were comprehensively reviewed. Data analysis included clinical presentation, computed tomography scan findings, injury site and trajectory, Glasgow Coma Scale and Glasgow Outcome Scale scores, and the St. Louis Scale. A total of 29 pediatric patients (79% male, 21% female, median age 8years) with GSWTH were analyzed. The mortality rate was 17%. A transventricular wound trajectory was found in 17% of patients and was associated with a poor prognosis (P < 0.001). The Glasgow Coma Scale showed insufficient specificity in injury severity assessment. A multiple logistic regression model predicting injury severity based on bullet type, gender, and time to admission had an accuracy of 80%, while a Decision Tree model improved accuracy to 96.6%. A stacking model combining multiple logistic regression and Decision Tree increased sensitivity to 87.5% and explained 65.5% of the variance. The findings emphasize the importance of a multifactorial approach in children with GSWTH, highlighting its effectiveness for precise outcome prediction.

Multi-Level Machine Learning Model to Improve the Effectiveness of Predicting Customers Churn Banks

Abstract This study presents a novel multi-level Stacking model designed to enhance the accuracy of customer churn prediction in the banking sector, a critical aspect for improving customer retention. Our approach integrates four distinct machine-learning algorithms – K-Nearest Neighbor (KNN), XGBoost, Random Forest (RF), and Support Vector Machine (SVM) – at the first level (Level 0). These algorithms generate initial predictions, which are then combined and fed into higher-level models (Level 1) comprising Logistic Regression, Recurrent Neural Network (RNN), and Deep Neural Network (DNN). We evaluated the model through three scenarios: Scenario 1 uses Logistic Regression at Level 1, Scenario 2 employs a Deep Convolutional Neural Network (DNN), and Scenario 3 utilizes a Deep Recurrent Neural Network (RNN). Our experiments on multiple datasets demonstrate significant improvements over traditional methods. In particular, Scenario 1 achieved an accuracy of 91.08%, a ROC-AUC of 98%, and an AUC-PR of 98.15%. Comparisons with existing research further underscore the enhanced performance of our proposed model.

A measurement model of pedestrian tolerance time under signal-controlled conditions

The tolerance of pedestrians to red light signals is a crucial factor in urban residents’ travel experiences and road traffic safety. To address the current lack of research on methods for measuring pedestrian red light tolerance time at crossings, this paper proposes a stacking model for predicting the red light tolerance time of individual pedestrians. The model employs the XGBoost (XGB), random forest (RF), support vector machine regression (SVR), and multilayer perceptron (MLP) models as primary models, with a multiple linear regression model as the secondary-layer meta-model, as well as Bayesian hyperparameter tuning. Using random survival forest (RSF) and K-means clustering methods, the dataset was divided into three categories: the low, medium, and high tolerance groups. The feature variables influencing the red light tolerance time were grouped accordingly. Stacking models were established for each tolerance and feature group. The experimental results demonstrated that the feature group voted on by multiple machine learning models performed the best in all three tolerance groups. In the low tolerance group, the mean squared error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) were 6.58, 1.91, and 19.78%, respectively. For the medium tolerant group, the MSE, MAE, and MAPE were 4.82, 1.53, and 7.63%, respectively. For the high tolerance group, the MSE, MAE, and MAPE were 33.32, 3.89, and 10.14%, respectively. Individual XGB, RF, SVR, MLP, and ungrouped stacking models were established for comparative analysis of the test set. The results indicated that the proposed grouped stacking model outperformed the other models overall. The method provides a means of obtaining the probability distribution function of the length of the waiting crowd’s tolerance time for each red light range and thus determining the duration of the red light at the crossing signal for different personnel compositions.

Mapping high-resolution XCO2 concentrations in China from 2015 to 2020 based on spatiotemporal ensemble learning model

High-resolution column-averaged dry air mole fraction of CO2 (XCO2) data is crucial for understanding the spatiotemporal patterns of XCO2 and for mitigating carbon emissions. Due to the limited scanning range of sensors and strict inversion conditions, satellite-retrieved XCO2 data are often significantly incomplete. Machine learning models are widely used to fill these gaps in satellite XCO2 data. However, the limitations of individual machine learning models and the complexity of the spatial distribution of XCO₂ mean that the accuracy of XCO2 predictions still needs improvement. In this study, a new spatiotemporal stacked ensemble learning model (STEL) was developed by combining random forest (RF), extremely randomized trees (ERT), extreme gradient boosting (XGBoost), optical gradient boosting (LightGBM), and categorical boosting (CatBoost) using the stacking ensemble learning methodology. Considering the spatiotemporal heterogeneity of XCO2, a novel spatiotemporal weighting feature was constructed as part of the model's input parameters. Finally, the XCO2 observed by Orbiting Carbon Observatory 2 (OCO-2) was reconstructed using STEL, and a monthly mean XCO2 dataset covering China from 2015 to 2020 was generated at a spatial resolution of 0.1°. The results show that STEL exhibits superior performance and generalization capabilities compared to individual machine-learning models. R2 RMSE and MAPE were 0.9624, 1.0023 ppm, and 0.1583 % on the test set, and 0.8970, 1.4213 ppm, and 0.2475 % for R2, RMSE, and MAPE in ground validation, respectively. In 10-fold cross-validation, STEL's RMSE was reduced by 9.52 % compared to the best-performing single model (RF). The spatiotemporal trend of CO2 in China from 2015 to 2020 was analyzed using STEL XCO2 data. The results indicate that this dataset accurately reflects the spatiotemporal heterogeneity of XCO2 distribution at a fine scale. Overall, XCO2 exhibited a spatiotemporal pattern of “high in the east and low in the west” and “high in spring and low in summer.” Except in summer, high XCO₂ values were mainly distributed in the North China Plain. XCO2 trends and hotspots showed considerable spatial variation. The Pearl River Delta and Yangtze River Delta urban agglomerations have the fastest XCO2 growth rates, and the distribution of XCO2 hotspots is consistent with the distribution of population and economic centers. In the sparsely populated northwest of China, XCO2 is growing rapidly due to increased thermal power generation and coal mining. XCO2 hotspots in Northwest China are mainly located in Xinjiang, Ningxia, and Inner Mongolia. The methodology and data presented are useful for further research on carbon emissions, carbon sinks, and climate change.

Comprehensive assessment of the water environment carrying capacity based on machine learning

Human activities and climate change are constantly threatening water environment systems. To alleviate pressure on the water environment and meet the requirements for sustainable societal development, it has become important to study the water environment carrying capacity (WECC). As science and technology have advanced, the use of artificial intelligence has penetrated various disciplines, and the application of machine learning has achieved many good results in the environmental sciences; however, little attention has been given to the WECC. To this end, a WECC assessment model was developed based on a variety of machine learning algorithms, to provide a novel approach and better understanding of the WECC. In this study, 13 natural and anthropogenic factors related to the water environment system in Liaoning Province (China) were selected, and the study period ranged from 2015 to 2019. Four basic machine learning algorithms, namely, random forest (RF), support vector machine (SVM), multilayer perceptron (MLP) and k-nearest neighbour (KNN), were used for batch modelling, and stacking was carried out based on the output results. A comparison and screening of the models revealed that RF and the stacking model had the best model prediction performances. Because RF was relatively simple and convenient, a comprehensive assessment of the regional WECC based on this algorithm was carried out. The results showed that most of the control units with an overloaded WECC are located in the central and northern regions of Liaoning Province, and the control units of overloading can be well identified and the trend of regional WECC status can be observed. To improve model transparency, we also performed an explanatory analysis based on the shapley additive explanations (SHAP). The results showed that precipitation, soil texture, sewage treatment plant, wind speed, and temperature were the most important factors influencing the water environment system. Precipitation is the most influential factor, and with increasing rainfall, the WECC first improves and then gradually deteriorates. For the anthropogenic factors, the WECC generally improves first and then worsens as these factors increase. These research results can be used to provide a powerful and practical new methodology for the comprehensive assessment of the WECC and provide a basis for the development of water environment policies and regulatory implementation in the region.

Monitor and classify dough based on color image with deep learning

To address subjective, time-consuming, and labor-intensive manual methods for monitoring and classifying fermenting dough, an automated and non-destructive method is proposed. Deep learning model YOLOv8s and extracted features including dough surface area, contrast, and homogeneity from RGB color images were employed to monitor fermenting dough. The features were input to a stacked ensemble model (SEM) with base models SVM, AdaBoost, KNN, and RF, with AdaBoost as meta-learner to classify fermenting dough into under-fermented, fermented, and over-fermented. SEM demonstrated a high dough classification rate of 83%, with specific rates of 75% for under-fermented, 71% for fermented, and 90% for over-fermented dough. Results reviewed that combining dough surface area and texture features is effective for monitoring dough, and can be used in adjusting chamber conditions. Furthermore, SEM showed great ability in classifying fermenting dough. The proposed method offers a promising solution for improved bread quality and consistency in bread-making.

Identification and Analysis of Credit Card Fraud Based on Machine Learning Methods

Credit card has become an indispensable payment tool in People's Daily life. However, credit card fraud has also increased, resulting in huge financial losses for banks and individuals. Based on the real world credit card transaction data, this paper uses a variety of machine learning algorithms to identify and analyze credit card fraud. Firstly, the problem of missing and unbalanced data is solved through data preprocessing and feature selection, and important features are screened out, Such as transaction amount, product code, payment card type, etc. Then, integrated logistic regression, random forest, LightGBM, XGBoost, and Stacking models were constructed, and the models were trained and evaluated through cross-validation and parameter optimization. The experimental results show that the integrated Stacking model performs best in credit card fraud detection, with an AUC(Area Under the Curve) value of 0.960, which can accurately identify fraudulent transactions and provide an effective fraud warning mechanism for financial institutions. This mechanism significantly benefits both financial institutions and credit card users by reducing fraud losses and enhancing transaction security. However, it is acknowledged that the study has limitations, including the potential impact of data imbalance on model performance, the need for further testing of model generalization ability to new fraud patterns, and the importance of optimizing real-time performance to ensure practical applicability.

A novel stacking ensemble learner for predicting residual strength of corroded pipelines

Accurately assessing the residual strength of corroded oil and gas pipelines is crucial for ensuring their safe and stable operation. Machine learning techniques have shown promise in addressing this challenge due to their ability to handle complex, non-linear relationships in data. Unlike previous studies that primarily focused on enhancing prediction accuracy through the optimization of single models, this work shifts the emphasis to a different approach: stacking ensemble learning. This study applies a stacking model composed of seven base learners and three meta-learners to predict the residual strength of pipelines using a dataset of 453 instances. Automated hyperparameter tuning libraries are utilized to search for optimal hyperparameters. By evaluating various combinations of base learners and meta-learners, the optimal stacking configuration was determined. The results demonstrate that the stacking model, using k-nearest neighbors as the meta-learner alongside seven base learners, delivers the best predictive performance, with a coefficient of determination of 0.959. Compared to individual models, the stacking model also significantly improves generalization performance. However, the stacking model’s effectiveness on low-strength pipelines is limited due to the small sample size. Furthermore, incorporating original features into the second-layer model did not significantly enhance performance, likely because the first-layer model had already extracted most of the critical features. Given the marginal contribution of model optimization to prediction accuracy, this work offers a novel perspective for improving model performance. The findings have important practical implications for the integrity assessment of corroded pipelines.

Predicting mechanical properties of sustainable green concrete using novel machine learning: Stacking and gene expression programming

Abstract Using rice husk ash (RHA) as a cement substitute in concrete production has potential benefits, including cement consumption and mitigating environmental effects. The feasibility of RHA on concrete strength was investigated in this research by predicting the split tensile strength (SPT) and flexural strength (FS) of RHA concrete (RHAC). The study used machine learning (ML) methods such as ensemble stacking and gene expression programming (GEP). The stacking model was improved using base learner configurations ML models, such as, random forest (RF), support vector regression, and gradient boosting regression. The proposed models were validated by statistical tests and external validation criteria. Moreover, the effect of input parameters was investigated using Shapley adaptive exPlanations (SHAP) for RF and parametric analysis for GEP-based models. The analysis revealed that the stacking ensemble integrates base learner predictions and demonstrated superior performance, with R values greater than 0.98 and 0.96. Mean absolute error and root mean square error values for both SPT and FS were 0.23, 0.3, 0.5, and 0.7 MPA, respectively. The SHAP analysis demonstrated water, cement, superplasticizer, and age as influential parameters for the RHAC strength. Furthermore, the SPT and FS of RHAC can be predicted with an acceptable error using the GEP expressions in the standard design procedure.

Ensemble machine learning models for predicting the CO2 footprint of GGBFS-based geopolymer concrete

While geopolymer concrete (GPC) has gained popularity for its environmentally friendly attributes compared to ordinary Portland cement, the absence of a prediction model for the carbon footprint of its constituents presents challenges for optimization within the evolving concrete industry. This study offers a thorough CO2 footprint prediction for ground granulated blast-furnace slag (GGBFS)-based GPC, utilizing advanced AI techniques, including a combination of machine learning models and stacking ensembles. This research statistically examines crucial parameters responsible for CO2 emissions in GGBFS-based GPC production, identifying factors like superplasticizer content, initial curing temperature, and NaOH (dry) content as significant contributors. Emphasizing sustainability, the study advocates optimizing concrete mixtures by considering factors like the NaOH ratio to other activator materials. After rigorously evaluating 12 machine learning models, including ensemble techniques, this study identified M4—a stacking model of Support Vector Regression (SVR) and Neural Network (NN)—as weak models, and Decision Tree (DT) as a meta-model, as the most effective ensemble model for predicting CO2 footprints. The choice of M4 is supported by various performance metrics such as the lowest Mean Squared Error of 88.8 and Root Mean Squared Error of 9.42, alongside the highest R2, Adjusted R2, and Explained Variance scores, all approximately 0.95. Additional analyses, such as Euclidean distance and Taylor diagrams, further substantiate the selection of M4. The findings have practical implications for sustainable and cleaner concrete production, enabling businesses to optimize the CO2 footprint of GGBFS-based GPC.

Stack-HDAC3i: A high-precision identification of HDAC3 inhibitors by exploiting a stacked ensemble-learning framework

Epigenetics involves reversible modifications in gene expression without altering the genetic code itself. Among these modifications, histone deacetylases (HDACs) play a key role by removing acetyl groups from lysine residues on histones. Overexpression of HDACs is linked to the proliferation and survival of tumor cells. To combat this, HDAC inhibitors (HDACi) are commonly used in cancer treatments. However, pan-HDAC inhibition can lead to numerous side effects. Therefore, isoform-selective HDAC inhibitors, such as HDAC3i, could be advantageous for treating various medical conditions while minimizing off-target effects. To date, computational approaches that use only the SMILES notation without any experimental evidence have become increasingly popular and necessary for the initial discovery of novel potential therapeutic drugs. In this study, we develop an innovative and high-precision stacked-ensemble framework, called Stack-HDAC3i, which can directly identify HDAC3i using only the SMILES notation. Using an up-to-date benchmark dataset, we first employed both molecular descriptors and Mol2Vec embeddings to generate feature representations that cover multi-view information embedded in HDAC3i, such as structural and contextual information. Subsequently, these feature representations were used to train baseline models using nine popular ML algorithms. Finally, the probabilistic features derived from the selected baseline models were fused to construct the final stacked model. Both cross-validation and independent tests showed that Stack-HDAC3i is a high-accuracy prediction model with great generalization ability for identifying HDAC3i. Furthermore, in the independent test, Stack-HDAC3i achieved an accuracy of 0.926 and Matthew’s correlation coefficient of 0.850, which are 0.44–6.11% and 0.83–11.90% higher than its constituent baseline models, respectively.

Stacked machine learning approach for predicting evolved hydrogen from sugar industry wastewater

The wastewater generated by the sugar industry contains high levels of organic pollutants, posing significant environmental and health risks. Traditional treatment methods often fall short in effectively managing these pollutants, necessitating innovative approaches for wastewater treatment and hydrogen production. In this study, machine learning (ML) algorithms are explored to predict hydrogen evolution from sugar industry wastewater treated via photocatalytic oxidation using solar-driven photocatalysts. By leveraging ML-driven models, we aim to streamline experimental procedures, reduce costs, and enhance energy efficiency. Our proposed stacking model, which combines Random Forest, Gradient Boosting, XGBoost, and Extra Trees, achieved superior results with R2 values of 0.9933, 0.9866, and 0.9906, and mean squared errors (MSE) of 0.0350, 0.0412, and 0.0412 for GLFO, GBFO, and GLRO, respectively. These results demonstrate the significant advantages of using ensemble techniques, particularly our proposed stacking method, which outperformed all individual models. The blending model, which averaged predictions from XGBoost and Gradient Boosting, also performed well with R2 values of 0.9866, 0.9811, and 0.9820, and MSEs of 0.0702, 0.0815, and 0.0795 for GLFO, GBFO, and GLRO, respectively. This study highlights the efficacy of ensemble methods in predicting hydrogen production, providing a novel and effective approach for wastewater treatment and renewable energy generation.

Urban air quality index forecasting using multivariate convolutional neural network based customized stacked long short-term memory model

In the context of increasing urban pollution and its adverse health effects, this study focuses on enhancing the air quality prediction model by forecasting future concentrations of various Air Quality Index (AQI) levels to support the development of green smart cities. The primary objective of this study is to develop a robust multivariate time series forecasting model using a combination of Convolutional Neural Networks (CNN) and Customized Stacked-based Long Short-Term Memory (CSLSTM) networks, namely C2SLSTM. The proposed methodology involved preprocessing AQI and meteorological data, scaling the features, and employing a hybrid CNN-LSTM architecture to capture spatial and temporal dependencies in the data. Dynamic AQI pattern changes are computed using online learning. The proposed model was trained on a big dataset containing AQIs such as particulate matter (PM2.5), carbon monoxide (CO), and nitrogen dioxide (NO2), along with meteorological factors like weather count, temperature, wind, humidity, and pressure. Experimental results demonstrate that the model achieved a mean absolute error (MAE) of 0.25, mean square error (MSE) of 0.083, root mean square error (RMSE) of 0.289 and a R2-Score of 0.84 on the test set. Also, compared to Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), Stacked Long Short-Term Memory (SLSTM), and Multilayer Perceptron (MLP) models, the C2SLSTM model showcases a performance enhancement of 35 %, 50 %, 30 %, and 32 %. Notably, the CNN component effectively extracted spatial features, while the LSTM layers captured temporal patterns, leading to precise predictions. These results indicate that the proposed approach effectively captures temporal dependencies and improves forecasting performance, with R2-Score values tending towards 1 for 12-hour prediction intervals. This research demonstrates the potential of advanced SLSTM architectures in accurately predicting AQI metrics, which can aid in better environmental monitoring and management. Also, the spatiotemporal correlation analysis is done through the designed ir metric.

A mechanical TiAl/TiAlN multinanolayer coating model based on microstructural analysis and nanoindentation

A finite element model reproducing the mechanical behaviour during nanoindentation tests on Ti0.67Al0.33/Ti0.54Al0.46N multilayer coatings was developed. Coatings with two different nanoperiods (10 and 50 nm) were deposited by radio-frequency magnetron sputtering from a single sintered titanium/aluminium target using the reactive gas pulsing process. X-ray diffraction and transmission electron microscopy confirmed the multilayer stacking of the coatings. The finite element models of coatings with these two stacking periods were built considering successive hypotheses: equal thicknesses for metal and nitride nanolayers stacked without transition layer, equal thicknesses stacking with a transition layer between each metal and nitride nanolayer and finally imbalanced thicknesses stacking with a transition layer. The elastic and plastic properties of the stacked nanolayers were determined on thick monolithic coatings of metal and nitride using indentation testing and the finite element model updating method respectively. The elastic-plastic properties of the interface were introduced in the multilayer model as a rule of mixtures of the metal and nitride properties using two hypotheses: parallel or serial. For 50 nm-period film, the interface has a negligible effect on the overall indentation response. On the other hand, for the 10 nm period, the imbalanced stacking model with precise knowledge of the transition layer thickness obtained by N K-edge electron energy loss spectroscopy is required to reproduce the experimental indentation curve. Compared to classical analytical models accounting for hardness, not only the hardnesses but also the indentation moduli appear to be well predicted and evaluated in this work.

Safety evaluation method for operational shield tunnels based on semi-supervised learning and a stacking algorithm

The safety assessment of structural defects in operational shield tunnels is crucial for ensuring their serviceability and safe operation. This study developed a novel comprehensive evaluation method for tunnel safety assessment based on semi-supervised learning and a stacking ensemble algorithm. First, in the membership degree calculation of the multidimensional normal cloud model, the importance coefficients of the tunnel safety evaluation indicators were used instead of their weights to improve the cloud model. This allowed generating unlabeled samples with patterns consistent with those of the collected samples with structural defects. Three classifiers, namely, random forest, extra trees, and gradient boosting, were employed for semi-supervised learning. This process converted unlabeled samples into pseudo-labeled samples, expanded the structural defects database, and ultimately optimized the classifier performance. Subsequently, a stacking algorithm was used to integrate the three optimized classifiers. This resulted in the creation of three stacking models, each containing multiple metamodels. Finally, the optimal metamodels were selected based on accuracy, precision, recall, and F1 score for a voting scheme to determine the tunnel safety level. The developed evaluation method was applied to a real-world engineering project. The evaluation results demonstrated consistency with those obtained from actual field conditions, thus validating the reliability and rationality of the proposed method.