Ensemble Prediction Model Research Articles

Association between long-term PM2.5 exposure and risk of Kawasaki disease in children: A nationwide longitudinal cohort study

BackgroundBased on previous studies suggesting air pollution as a potential risk factor for Kawasaki Disease (KD), we examined the association of long-term exposure to childhood fine particulate matter (PM2.5) with the risk of KD. MethodsWe used National Health Insurance Service–National Sample Cohort data from 2002 to 2019, which included beneficiaries aged 0 years at enrollment and followed-up until the onset of KD or age 5 years. The onset of KD was defined as the first hospital visit record with a primary diagnostic code of M30.3, based on the 10th revision of the International Classification of Diseases, and with an intravenous immunoglobulin (IVIG) prescription. We assigned PM2.5 concentrations to 226 districts, based on mean annual predictions from a machine learning-based ensemble prediction model. We performed Cox proportional-hazards modeling with time-varying exposures and confounders. ResultsWe identified 134,634 individuals aged five or less at enrollment and, of these, 1220 individuals who had a KD onset and an IVIG prescription during study period. The average annual concentration of PM2.5 exposed to the entire cohort was 28.2 μg/m³ (Standard Deviation 2.9). For each 5 μg/m³ increase in annual PM2.5 concentration, the hazard ratio of KD was 1.21 (95% CI 1.05–1.39). ConclusionsIn this nationwide, population-based, cohort study, long-term childhood exposure to PM2.5 was associated with an increased incidence of KD in children. The study highlights plausible mechanisms for the association between PM2.5 and KD, but further studies are needed to confirm our findings.

An ensemble model for monthly runoff prediction using least squares support vector machine based on variational modal decomposition with dung beetle optimization algorithm and error correction strategy

In order to enhance the runoff prediction accuracy, an ensemble prediction model based on least squares support vector machine (LSSVM) is proposed by including variational mode decomposition (VMD), dung beetle optimization algorithm (DBO), and error correction (EC) strategy. First, the monthly runoff time series is decomposed using DBO-optimized VMD (DVMD), yielding a series of intrinsic mode functions (IMF) series and a residual (Res). Then, the LSSVM based on DBO optimization predicts each sub-series column and residual. The final forecast results are achieved after the preliminary forecast results have been stacked and corrected by the DBO-LSSVM prediction error. To verify the reliability of the proposed model, it is applied to the monthly runoff prediction of the Xiajiang hydrological station in the Ganjiang River Basin, the Hongshanhe hydrological station in the Heihe River Basin, and the Jiayugaun hydrological station in the Heihe River Basin. The proposed model is evaluated using four evaluation indicators: RMSE, MAPE, NSEC, and R, and is compared with SVM, LSSVM, PSO-LSSVM, DBO-LSSVM, EEMD-LSSVM, CEEMDAN-LSSVM, DVMD-LSSVM, EEMD-DBO-LSSVM, CEEMDAN-DBO-LSSVM, and DVMD-DBO-LSSVM. Results show that the DVMD-DBO-LSSVM-EC model has the highest accuracy. During the test period, the NSEC of Xiajiang hydrological station is 0.9829, R is 0.9921, the NSEC of Hongshanhe hydrological station is 0.9981, R is 0.9991, and the NSEC of Jiayugaun hydrological station is 0.9772, R is 0.9897. The prediction effect of the model on the extreme value of the three stations after adding the error correction strategy has increased by 45.14%, 62.22%, and 29.49%, respectively, compared with the previous, which is closer to the actual value. The developed combination model offers a new approach to forecasting monthly runoff and extreme values.

Ensemble model for accuracy prediction of protein secondary structure

<span>Predicting a protein’s secondary structure is crucial for understanding the working of proteins. Despite advancements over the years, the top predictors have achieved only 80% Q8 accuracy when sequence profile information is the sole input. An ensemble approach is proposed using convolutional neural network (CNN) and a classifier known as support vector machine (SVM) on both the partial and the whole CullPDB datasets. The protein secondary structure (PSS) has a complex hierarchical structure, as well as the ability to take into account the reliance between neighbouring labels. A detailed experiment yielding high levels of Q8 accuracy with scores of 97.91%, 85.13%, and 78.02% using 20%, 80%, and 100% respectively of the protein residues on the new predicted dataset CullPDB6133 which is better than the accuracies predicted by similar models. The proposed methodology highlights the use of CNN as a general framework, for efficiently predicting eight-state (Q8) accuracy of secondary protein structures with a low time and space complexity.</span>

Input-parameter optimization using a SVR based ensemble model to predict landslide displacements in a reservoir area – A comparative study

For reservoir landslides with limited in situ monitoring data, improving the accuracy of an ensemble prediction model through historical displacement-related factor deconstruction and model optimization is challenging. Unfortunately, even though various support vector regression (SVR) based ensemble prediction models have been proposed for this purpose, the optimal combination of new algorithms and their degree of improved prediction accuracy is still unclear. Based on four evaluation indicators, this paper presents a comparative study of a typical landslide displacement prediction model containing four important nodes (landslide displacement decomposition, inducing factor frequency-component extraction, input parameter selection, and optimization algorithm selection) vital to prediction accuracy. The empirical mode decomposition (EMD) series model was adopted to decompose the landslide displacement and extract the frequency component of the factors that affect the landslide movement. Four trajectory similarity judgment models were used to select input variables for the SVR-based prediction model. Nine swarm intelligence (SI) algorithms were used to help optimize the landslide prediction model. A case study shows that the SVR-based ensemble landslide prediction model works well in predicting the displacement of a slow-moving landslide. The combination of EEMD-CEEMDAN-LCSS-PSO yields the best prediction performance, with MAPE, RMSE, MAE, and R2 values of 0.004, 2.38, 8.64, and 0.96, respectively. The findings of this study can impact further studies of landslide ensemble prediction model optimization with insights into the algorithm selection, parameter setting, evaluation measures, and experimental settings.

Diagnosis of Cardiovascular Diseases by Ensemble Optimization Deep Learning Techniques

Cardiovascular disease (CVD) is a variety of diseases that affect the blood vessels and the heart. The authors propose a set of deep learning inspired by the approach used in CVD support centers for the early diagnosis of CVD using deep learning techniques. Data were collected from patients who received CVD screening. The authors propose a prediction model to diagnose whether people have CVD or not and to provide awareness or diagnosis on that. The performance of each algorithm is compared with that of long-, short-time memory, feedforward, and cascade forward neural networks, and Elman neural networks. The results show that the ensemble deep learning classification and prediction model achieved 98.45% accuracy. Using the proposed early diagnosis model for CVD can help simplify the diagnosis of CVD by medical professionals.

6mA-StackingCV: an improved stacking ensemble model for predicting DNA N6-methyladenine site

DNA N6-adenine methylation (N6-methyladenine, 6mA) plays a key regulating role in the cellular processes. Precisely recognizing 6mA sites is of importance to further explore its biological functions. Although there are many developed computational methods for 6mA site prediction over the past decades, there is a large root left to improve. We presented a cross validation-based stacking ensemble model for 6mA site prediction, called 6mA-StackingCV. The 6mA-StackingCV is a type of meta-learning algorithm, which uses output of cross validation as input to the final classifier. The 6mA-StackingCV reached the state of the art performances in the Rosaceae independent test. Extensive tests demonstrated the stability and the flexibility of the 6mA-StackingCV. We implemented the 6mA-StackingCV as a user-friendly web application, which allows one to restrictively choose representations or learning algorithms. This application is freely available at http://www.biolscience.cn/6mA-stackingCV/. The source code and experimental data is available at https://github.com/Xiaohong-source/6mA-stackingCV.

An Ensemble Learning-Enhanced Smart Prediction Model for Financial Credit Risks

The credit risk assessment acts as an important part in daily affairs for financial institutions. But in the era of big data, the growing business volume makes it an urgent demand to develop digital ways of credit risk assessment. Currently, the machine learning is universally employed to establish various data-driven models for this purpose. However, machine learning models generally suffer from limited ability of feature representation and robustness, and cannot deal with more complex financial security scenarios. To deal with this issue, this work introduces ensemble learning to construct a stronger credit risk prediction model via integration of several basic machine learning models. Thus, an ensemble learning-enhanced smart prediction model for financial credit risk is proposed in this paper. Three classification-based machine learning models (support vector machine, artificial neural network and radial basis function) are selected as the basic classifiers, and “voting” strategy is utilized to integrate them into a novel strong classifier. A real-world financial credit dataset released by a Chinese commercial bank was selected as the experimental scenario. The obtained results show that the proposal has better prediction accuracy compared with basic machine learning models without ensemble learning.

Imaging-Based Biomarkers Predict Programmed Death-Ligand 1 and Survival Outcomes in Advanced NSCLC Treated With Nivolumab and Pembrolizumab: A Multi-Institutional Study

BackgroundAlthough the immune checkpoint inhibitors, nivolumab and pembrolizumab, were found to be promising in patients with advanced NSCLC, some of them either do not respond or have recurrence after an initial response. It is still unclear who will benefit from these therapies, and, hence, there is an unmet clinical need to build robust biomarkers. MethodsPatients with advanced NSCLC (N = 323) who were treated with pembrolizumab or nivolumab were retrospectively identified from two institutions. Radiomics features extracted from baseline pretreatment computed tomography scans along with the clinical variables were used to build the predictive models for overall survival (OS), progression-free survival (PFS), and programmed death-ligand 1 (PD-L1). To develop the imaging and integrative clinical-imaging predictive models, we used the XGBoost learning algorithm with ReliefF feature selection method and validated them in an independent cohort. The concordance index for OS, PFS, and area under the curve for PD-L1 was used to evaluate model performance. ResultsWe developed radiomics and the ensemble radiomics-clinical predictive models for OS, PFS, and PD-L1 expression. The concordance indices of the radiomics model were 0.60 and 0.61 for predicting OS and PFS and area under the curve was 0.61 for predicting PD-L1 in the validation cohort, respectively. The combined radiomics-clinical model resulted in higher performance with 0.65, 0.63, and 0.68 to predict OS, PFS, and PD-L1 in the validation cohort, respectively. ConclusionsWe found that pretreatment computed tomography imaging along with clinical data can aid as predictive biomarkers for PD-L1 and survival end points. These imaging-driven approaches may prove useful to expand the therapeutic options for nonresponders and improve the selection of patients who would benefit from immune checkpoint inhibitors.

An ensemble learning-based experimental framework for smart landslide detection, monitoring, prediction, and warning in IoT-cloud environment.

Landslides occur every year during the monsoon season in hilly areas. This natural disaster annually leads to several fatalities, injuries, and property destruction. Monitoring landslides and promptly alerting people to looming disasters in light of these injuries and fatalities are crucial. To date, no efficient technique is in practice to predict landslides. The tools that are now available monitor landslides at a very high cost and do not offer early warning or forecasts of soil movement. An innovative, low-cost Internet of Things (IoT)-based system for landslip warning, monitoring, and prediction is the major objective of this research. Its assessment, implementation, and development are described in detail. This study proposes an IoT-based smart landslide detection, warning, prediction, and monitoring system. The pre and post-measures use sensors and other hardware to deal with landslide disasters. It uses real-time environment monitoring (landslide site) for any changes and provides appropriate output by comparing the threshold values. The proposed system is tested on a prototype model, which performed well in our tests. The database was updated 2.5 s after the landslide thanks to a steady Internet connection. In less than 5 s after the event, the Thingspeak channel can display a graphical depiction of the data and its position. Multiple readings showed an 80-85% system accuracy rate. Further, the proposed ensemble learning-based risk prediction model is applied to static and dynamic data to predict the landslide for future reference. The ensemble classifier model has 98.67% recall, 96.56% accuracy, 97.35% F1-value, and 96.07% precision. The alert SMS is also sent to concerned authorities for medical emergency/PWD department/district administration.

An ensembled remaining useful life prediction method with data fusion and stage division

The remaining useful life (RUL) prediction method based on multi-sensor vibration data is a significant component of predictive maintenance for rolling bearings. However, during the fusion process, it is easy to overlook the consistency of multi-sensor vibration data and cannot adaptively divide degradation stages, resulting in a decrease in the accuracy of the prediction method and limits its applicability in industrial settings. Therefore, this article proposes an integrated prediction method for the RUL of rolling bearings based on data fusion and stage division. Firstly, a data-level fusion method based on multi-sensor vibration signals (MSDF) is proposed. This method dynamically weights sensor data, aiming to consider consistency and reliability in order to achieve data level fusion for multi-sensor vibration signals. Secondly, a stage division method is proposed, which adaptively divides the degradation process into three stages to guide data fusion and ensemble prediction results. Finally, the feature complementarity based ensemble prediction (TCEP) model is proposed to enhance prediction accuracy by learning the degradation difference information of features throughout the prediction process. Furthermore, the outstanding performance of the proposed method was validated using two sets of bearing lifetime vibration signal datasets.

Improving the Landslide Susceptibility Prediction Accuracy by Using Genetic Algorithm Optimized Machine Learning Approach

Landslide susceptibility prediction is critical in open pit mines and geotechnical fields. Prediction accuracy is very essential to reduce the risk of slope instability. Traditional statistical learning methods have been widely used in early warning systems, but they cannot thoroughly explore the coupling effect among related factors, which often results in low prediction accuracy. This paper establishes an ensemble learning prediction model optimized by a genetic algorithm (GA) to determine landslide susceptibility more quickly and efficiently. The model is based on 290 sets of slope cases containing height (H), slope angle (α), unit weight (γ), cohesion (c), friction angle (φ), and pore water pressure (ru). Two common algorithms are incorporated into the ensemble learning model: Xgboost and gradient boosting decision tree (GBDT). The area under the curve (AUC) of GA-GBDT and GA-Xgboost were found to be 0.928 and 0.933, respectively, both of which could predict landslide susceptibility better. Compared with multiple logistic regression and other machine learning algorithms, both GA-GBDT and GA-Xgboost models perform better in terms of accuracy and applicability. The study results demonstrate that the developed optimized machine learning model can accurately predict landslide susceptibility and that the parameters should be optimized on a case-by-case basis to achieve more accurate results after building a suitable machine model. The optimization model proposed in this paper can be an effective new method for the intelligent prediction of landslide susceptibility.

Ensemble of ensembles for fine particulate matter pollution prediction using big data analytics and IoT emission sensors

Purpose The study aims to develop a multilayer high-effective ensemble of ensembles predictive model (stacking ensemble) using several hyperparameter optimized ensemble machine learning (ML) methods (bagging and boosting ensembles) trained with high-volume data points retrieved from Internet of Things (IoT) emission sensors, time-corresponding meteorology and traffic data. Design/methodology/approach For a start, the study experimented big data hypothesis theory by developing sample ensemble predictive models on different data sample sizes and compared their results. Second, it developed a standalone model and several bagging and boosting ensemble models and compared their results. Finally, it used the best performing bagging and boosting predictive models as input estimators to develop a novel multilayer high-effective stacking ensemble predictive model. Findings Results proved data size to be one of the main determinants to ensemble ML predictive power. Second, it proved that, as compared to using a single algorithm, the cumulative result from ensemble ML algorithms is usually always better in terms of predicted accuracy. Finally, it proved stacking ensemble to be a better model for predicting PM2.5 concentration level than bagging and boosting ensemble models. Research limitations/implications A limitation of this study is the trade-off between performance of this novel model and the computational time required to train it. Whether this gap can be closed remains an open research question. As a result, future research should attempt to close this gap. Also, future studies can integrate this novel model to a personal air quality messaging system to inform public of pollution levels and improve public access to air quality forecast. Practical implications The outcome of this study will aid the public to proactively identify highly polluted areas thus potentially reducing pollution-associated/ triggered COVID-19 (and other lung diseases) deaths/ complications/ transmission by encouraging avoidance behavior and support informed decision to lock down by government bodies when integrated into an air pollution monitoring system Originality/value This study fills a gap in literature by providing a justification for selecting appropriate ensemble ML algorithms for PM2.5 concentration level predictive modeling. Second, it contributes to the big data hypothesis theory, which suggests that data size is one of the most important factors of ML predictive capability. Third, it supports the premise that when using ensemble ML algorithms, the cumulative output is usually always better in terms of predicted accuracy than using a single algorithm. Finally developing a novel multilayer high-performant hyperparameter optimized ensemble of ensembles predictive model that can accurately predict PM2.5 concentration levels with improved model interpretability and enhanced generalizability, as well as the provision of a novel databank of historic pollution data from IoT emission sensors that can be purchased for research, consultancy and policymaking.

A novel multi-layer stacking ensemble wind power prediction model under Tensorflow deep learning framework considering feature enhancement and data hierarchy processing

Wind power prediction is crucial for energy production, but due to the complicated data characteristics of wind farms, it's difficult to accurately predict wind power output and it is challenging for a single prediction model to properly handle multi-featured time-series data sets. To solve the above problem, the density-based spatial clustering of applications with noise (DBSCAN) method was used to find outliers of data sets, and the recursive feature elimination (RFE) method was used to screen out main features according to importance. Sequentially a multi-layer stacking ensemble learning prediction model was constructed based on data hierarchy processing and feature enhancement techniques. Seven groups of ablation experiments were set to verify the model's effectiveness with the annual time-series data. The experimental results show that the DBSCAN method can effectively find outliers in wind data sets and improve the prediction accuracy and the RFE method can significantly reduce the computing time and improves the generalization ability and prediction accuracy. The multi-layer stacking ensemble model can slow down the overfitting trend of a single prediction model, effectively avoid the model from falling into the local search state, and obtain a strong learner with better generalization ability.

Abstract 17098: Detection of Multiple Structural Heart Diseases Using a Novel Artificial Intelligence-Driven Algorithm for Noisy Real-World Single-Lead ECGs

Background: Wearable devices capable of ECG acquisition can help scale screening of structural heart disease (SHD) if diagnoses can be made with a single lead ECG. Algorithms for portable and wearable ECG require resilience to noisy acquisition but can enable detecting SHD in a community setting before symptom onset, ideal for public health management. Methods: Using Yale EHR data (2015-2021), we developed an artificial intelligence model to detect the presence of left ventricular systolic dysfunction, severe left-sided valvular disease, or hypertrophic cardiomyopathy using lead I of an ECG, which is commonly captured by wearable devices. We used 563,668 ECGs paired with echocardiograms obtained within 30 days of each other from 131,804 patients to develop the model. The dataset was split into training, validation, and test sets (85:5:10). For individual disease labels, we trained 6 convolutional neural network models using ECG signals augmented with random Gaussian real-world noise within 4 distinct frequency ranges and at 4 signal-to-noise ratios. Then, we included their prediction probabilities and age and sex as inputs to an XGBoost model to predict the composite SHD (Fig). Results: In the development set, the prevalence of composite SHD was 13%, encompassing left ventricular systolic dysfunction (9%), severe valvular disease (4%), and hypertrophic cardiomyopathy (1%). In the held-out test set of 13,181 patients (65±17 years, 49% women, 14% Black), AUROC and AUPRC of the ensemble model for prediction of SHD were 0.92 (0.90-0.93) and 0.61 (0.54-0.68), respectively (Fig). With optimized Youden's index, the model has a sensitivity of 0.83, specificity of 0.87, PPV of 0.41, and NPV of 0.98 to detect SHD. Conclusion: A novel wearable-adapted model demonstrated promising performance for detecting multiple SHD using single-lead ECG. This can enable community screening for SHD and improve patient outcomes through early initiation of guideline-directed therapies.

Ensemble deep learning enabled multi-condition generative design of aerial building machine considering uncertainties

The Aerial Building Machine (ABM) is a complex construction equipment used in high-rise building construction, facing challenges due to environmental uncertainties. This paper introduces a multi-condition generative design framework to improve ABM's visualization, operability, and intelligence. It seamlessly integrates real-time data between geometric and physical models and employs an ensemble deep learning model for objective value prediction, using a snapshot strategy. Combining structural reliability concepts with Latin hypercube sampling-based stochastic optimization, an optimal design scheme is obtained for uncertain loads. An ABM case study in China illustrates the approach's feasibility, showing it meets reliability requirements across different conditions and achieves significant improvements (16.59% under normal conditions and 16.91% under extreme wind conditions). Additionally, ensemble deep learning outperforms existing methods for ABM structural reliability estimation. Identifying optimal designs and evaluation options, this paper contributes a multi-condition optimization approach for enhanced structural reliability and establishes an efficient generative design workflow and system for exploring a vast solution space.

Interpretable ensemble prediction for anaerobic digestion performance of hydrothermal carbonization wastewater

Hydrothermal carbonization (HTC) is a method to improve fuel quality that can directly treat wet solid waste, but the treatment produces large amounts of wastewater. Hydrothermal carbonation wastewater treatment for methane production by anaerobic digestion can lead to waste utilization and energy saving. However, anaerobic digestion performance prediction of HTC wastewater is challenging due to the complexity of influencing factors. This study applies interpretable machine learning combined with ensemble learning to construct ensemble prediction models for the biogas yield and CH4 concentration. The machine learning ensemble model can integrate the advantages of single models and effectively improve the prediction accuracy of the anaerobic digestion performance of HTC wastewater, with the best R2 reaching 0.836 and 0.820, respectively, which is better than 0.780 and 0.802 of the best single models. The SHapley Additive exPlanations theory is combined with the ensemble models to show that anaerobic digestion reacted time with HTC temperature, pH, and COD has a coupling effect on daily biogas yield and CH4 concentration.

An Ensemble Forecast Wind Field Correction Model with Multiple Factors and Spatio-Temporal Features

Accurate wind speed prediction is significantly important for the full utilization of wind energy resources and the improvement in the economic benefits of wind farms. Because the ensemble forecast takes into account the uncertainty of information about the atmospheric motion, domestic and foreign weather service forecast centers often choose to use the ensemble numerical forecast to achieve the fine forecast of wind speed. However, due to the unavoidable systematic errors of the ensemble numerical forecast model, it is necessary to correct the deviation in the ensemble numerical forecast wind speed. Considering the typical spatio-temporal characteristics of the grid prediction data of the wind field, based on Convolutional Long–Short Term Memory (ConvLSTM) units and attention mechanism, this paper takes the complex and representative North China region as the research area, aiming to reveal the shortcomings of existing deep learning integrated prediction correction models in extracting temporal features of grid prediction data. We propose a new ensemble prediction wind field correction model integrating multi-factor and spatio-temporal characteristics. This model uses reanalyzed land data provided by the European Center for Medium-Range Weather Forecasts as the real data to correct the deviation in the near-surface 10 m wind field data predicted by the regional ensemble numerical prediction model of the China Meteorological Administration. We used the reanalyzed land data provided by the European Centre for Medium-Range Weather Forecasts (ECMWF) as the live data to correct the deviation in the near-surface 10 m wind field data predicted by the regional ensemble numerical forecast model of the China Meteorological Administration (CMA). At the same time, Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) were used as the scoring indicators, and the results of the China Meteorological Administration–Regional Ensemble Prediction System (CMA–REPS) ensemble average, multiple linear regression method correction, Long–Short Term Memory (LSTM) method correction, and U-net (UNET) method correction were compared. Compared with the UNET model method, the experimental results show that when processing the 10 m zonal wind data, 10 m meridional wind data, and 10 m average wind speed data of CMA–REPS 24 h forecasts, the correction results of our model can reduce the RMSE score index by 9.15%, 4.83%, and 7.79%. At the same time, when processing the 48 h and 72 h near-surface 10 m wind field data of the CMA–REPS forecast, our model can improve the prediction accuracy of CMA–REPS near-surface wind forecast data. Therefore, the correction effect of the proposed model in a complex terrain area is evidently better compared to other methods.

Evaluation of IPCC Models’ Performance in Simulating Late-Twentieth-Century Weather Patterns and Extreme Precipitation in Southeastern China

Climate model evaluation work has made progress both in theory and practice, providing strong support for better understanding and predicting climate change. However, at the weather scale, there is relatively little assessment of climate models in terms of daily-scale climate phenomena, such as storm frequency and intensity. These weather-scale variables are of significant importance for our understanding of the impacts of climate change. In order to assess the capability of climate models to simulate weather-scale climate patterns, this study employs Self-Organizing Maps (SOMs) for weather pattern classification. By combining different evaluation metrics, varying the number of SOM types, changing the size of the study area, and altering the reference datasets, the climate models are evaluated to ensure the robustness of the assessment results. The results demonstrate that the size of the study area is positively correlated with observed differences, and there are correlations among different evaluation metrics. The highest correlation is observed between evaluation metrics in large-scale and small-scale spatial domains, while the correlation with SOM size is relatively low. This suggests that the choice of evaluation metrics has a minor impact on model ranking. Furthermore, when comparing the correlation coefficients calculated using the same evaluation metrics for different-sized regions, a significant positive correlation is observed. This indicates that variations in the size of the study area do not significantly affect model ranking. Further investigation of the relationship between model performance and different SOM sizes shows a significant positive correlation. The impact of dataset selection on model ranking is also compared, revealing high consistency. This enhances the reliability of model ranking. Taking into account the influence of evaluation metric selection, SOM size, and reanalysis data selection on model performance assessment, significant variations in model ranking are observed. Based on cumulative ranking, the top five models identified are ACCESS1-0, GISS-E2-R, GFDL-CM3, MIROC4h, and GFDL-ESM2M. In conclusion, factors such as evaluation metric selection, study area size, and SOM size should be considered when assessing model ranking. Weather pattern classification plays a crucial role in climate model evaluation, as it helps us better understand model performance in different weather systems, assess their ability to simulate extreme weather events, and improve the design and evaluation methods of model ensemble predictions. These findings are of great significance for optimizing and strengthening climate model evaluation methods and provide valuable insights for future research.

Deep-Reinforcement-Learning-Based Dynamic Ensemble Model for Stock Prediction

In stock prediction problems, deep ensemble models are better adapted to dynamically changing stock market environments compared to single time-series networks. However, the existing ensemble models often underutilize real-time market feedback for effective supervision, and base models are pre-trained and fixed in their optimization, which makes them lack adaptability for evolving market environments. To address this issue, we propose a deep-reinforcement-learning-based dynamic ensemble model for stock prediction (DRL-DEM). Firstly, we employ deep reinforcement learning to optimize the weights of deep-learning-based time-series models. Secondly, existing deep-reinforcement-learning methods only consider environmental rewards. Thus we improve the reward function by introducing real-time investment returns as additional feedback signals for the deep-reinforcement-learning algorithm. Finally, an alternating iterative algorithm is used to simultaneously train the base predictors and the deep-reinforcement-learning model, allowing DRL-DEM to fully utilize the supervised information for global coordinated optimization. The experimental results show that in SSE 50 and NASDAQ 100 datasets, the mean square error (MSE) of the proposed method reached 0.011 and 0.005, the Sharpe ratio (SR) reached 2.20 and 1.53, and the cumulative return (CR) reached 1.38 and 1.21. Compared with the best results in the recent model, MSE decreased by 21.4% and 28.6%, SR increased by 81.8% and 82.1%, and CR increased by 89.0% and 89.1%, with higher forecasting accuracy and stronger investment return capability.

A bagging-based selective ensemble model for churn prediction on imbalanced data

A bagging-based selective ensemble model for churn prediction on imbalanced data