Ensemble Prediction Research Articles

Deep learning-based ensemble forecast and predictability analysis of the Kuroshio intrusion into the South China Sea

Abstract The Kuroshio intrusion (KI) into the South China Sea (SCS) significantly affects the environment, ecology, and climate change of the SCS. However, due to the nonlinearity of KI, its numerical prediction often requires large ensemble size to measure prediction uncertainty. The huge computational costs of large numbers of members and high-resolution numerical models pose significant challenges for KI prediction. We, therefore, construct a Kuroshio ensemble deep learning prediction system (KurNet) through taking different values of parameters to predict KI paths because the deep learning models have high prediction skills and low computational cost. The KurNet containing 64 ensemble members can not only output ensemble mean forecast result of the Kuroshio path, but also estimate probability density functions for the path types. The KurNet illustrates a high predictive ability for the KI with the mean classification accuracy of 71.9% and root mean square error of 0.913 on the testing set, which is superior to the single control prediction by ∼1.0–2.9%, although the control prediction model is selected as one of the ensemble members with the best predictive ability on the validation set. Furthermore, the predictability analysis of 10 KI events indicates that when the lead time is 3 days, the most important areas are in the east of Luzon Island due to the upstream Kuroshio transport. As the lead time increases, the most important area is in the Luzon Strait due to the eddy activity. Observing system simulation experiments reveal that the KI forecast skill can be enhanced by ∼12–18%, when uncertainties of the input data in these important regions are removed.

Application of Ensemble Algorithm Based on the Feature-Oriented Mean in Tropical Cyclone-Related Precipitation Forecasting

Tropical cyclones (TCs) are characterized by robust vortical motion and intense thermodynamic processes, often causing damage in coastal cities as they result in landfall. Accurately estimating the ensemble mean of TC precipitation is critical for forecasting and remains a foremost global challenge. In this study, we develop an ensemble algorithm based on the feature-oriented mean (FM) suitable for spatially discrete variables in precipitation ensembles. This method can adjust the locations of ensemble precipitation fields to reduce the location-related deviations among ensemble members, ultimately enhancing the ensemble mean forecast skill for TC precipitation. To evaluate the feasibility of the FM in TC precipitation ensemble forecasting, 18 landing TC cases in China from 2019 to 2021 were selected for validation. For precipitation forecasts of the landing TCs with a varying leading time, we conducted a comprehensive quantitative evaluation and comparison of the precipitation forecast skills of the FM and arithmetic mean (AM) algorithms. The results indicate that the field adjustment algorithm in the FM can effectively align with the TC precipitation structure and the location of the ensemble mean, reducing the spatial divergence among precipitation fields. The FM method demonstrates superior performance in the equitable threat score, probability of detection, and false alarm ratio compared with the AM, exhibiting an overall improvement of around 10%. Furthermore, the FM ensemble mean shows a higher pattern of the correlation coefficient with observations and has a smaller root mean square error than the AM ensemble mean, signifying that the FM method can better preserve the characteristics of the precipitation structure. Additionally, an object-based diagnostic evaluation method was used to verify forecast results, and the results suggest that the attribute distribution of FM forecast objects more closely resembles that of observed precipitation objects (including the area, longitudinal and latitudinal centroid locations, axis angle, and aspect ratio).

A novel hybrid supervised and unsupervised hierarchical ensemble for COVID-19 cases and mortality prediction

Though COVID-19 is no longer a pandemic but rather an endemic, the epidemiological situation related to the SARS-CoV-2 virus is developing at an alarming rate, impacting every corner of the world. The rapid escalation of the coronavirus has led to the scientific community engagement, continually seeking solutions to ensure the comfort and safety of society. Understanding the joint impact of medical and non-medical interventions on COVID-19 spread is essential for making public health decisions that control the pandemic. This paper introduces two novel hybrid machine-learning ensembles that combine supervised and unsupervised learning for COVID-19 data classification and regression. The study utilizes publicly available COVID-19 outbreak and potential predictive features in the USA dataset, which provides information related to the outbreak of COVID-19 disease in the US, including data from each of 3142 US counties from the beginning of the epidemic (January 2020) until June 2021. The developed hybrid hierarchical classifiers outperform single classification algorithms. The best-achieved performance metrics for the classification task were Accuracy = 0.912, ROC-AUC = 0.916, and F1-score = 0.916. The proposed hybrid hierarchical ensemble combining both supervised and unsupervised learning allows us to increase the accuracy of the regression task by 11% in terms of MSE, 29% in terms of the area under the ROC, and 43% in terms of the MPP metric. Thus, using the proposed approach, it is possible to predict the number of COVID-19 cases and deaths based on demographic, geographic, climatic, traffic, public health, social-distancing-policy adherence, and political characteristics with sufficiently high accuracy. The study reveals that virus pressure is the most important feature in COVID-19 spread for classification and regression analysis. Five other significant features were identified to have the most influence on COVID-19 spread. The combined ensembling approach introduced in this study can help policymakers design prevention and control measures to avoid or minimize public health threats in the future.

Construction of a Cutting-Tool Wear Prediction Model through Ensemble Learning

This study begins by conducting side milling experiments on SKD11 using tungsten carbide TiAlN-coated end mills to compare the surface roughness performance between two combinations of milling process parameters (feed rate and radial depth of cut), along with three ultrasonic-assisted methods (rotary, dual-axis, and rotary combined with dual-axis). The results suggest that the rotary (z-axis oscillation) ultrasonic-assisted method may provide better performance. Subsequently, this superior ultrasonic-assisted method was applied both with and without laser locally preheating assistance, respectively. Using a Taguchi orthogonal array, milling process parameters (spindle speed, feed rate, and radial depth of cut) were planned for experiments with the same cutting tool and the workpiece just mentioned above. The surface roughness serves as the objective function while being constrained by cutting-tool life. The characteristics of the smaller-the-better in the Taguchi method were applied to determine the optimal combination of process parameters. Based on the optimal milling process parameters obtained and the superior hybrid-assisted method adopted, milling experiments were repeatedly performed to collect the data on cutting force and cutting-tool wear. Feature engineering was performed on the cutting force signals, and different domain characteristics from both the time and frequency domains were extracted. Hereafter, feature selection by random forest and data standardization were further applied to feature extractions, and the data processing was thus completed. For the processed data, a cutting-tool wear prediction model was constructed by ensemble learning. This method leverages various machine learning regression models, including decision tree, random forest, extremely randomized tree, light gradient boosting machine, extreme gradient boosting, AdaBoost, stochastic gradient descent, support vector regression, linear support vector regression, and multilayer perceptron. After hyper-parameter tuning, the ensemble voting regression prediction was performed based on these ten mentioned models. The experimental results demonstrate that the ensemble voting regression model surpasses the performance of each individual machine learning regression model. In addition, this regression model achieves a coefficient of determination (R2) of 0.94576, a root mean square error (RMSE) of 0.24348, a mean squared error (MSE) of 0.05928, and a mean absolute error (MAE) of 0.18182. Therefore, the ensemble learning approach has been proven to be a feasible and effective method for monitoring cutting-tool wear.

Self-clustered GAN for precipitation nowcasting

This paper proposes a novel GAN framework with self-clustering approach for precipitation nowcasting (ClusterCast). Previous studies have primarily captured the motion vector using only a single latent space, making the models difficult to adapt to disparate space-time distribution of precipitation. Environmental factors (e.g., regional characteristics and precipitation scale) have an impact on precipitation systems and can cause non-stationary distribution. To tackle this problem, our key idea is to train a generator network to predict future radar frames by learning a sub-network that automatically labels precipitation types from a generative model. The training process consists of (i) clustering the hierarchical features derived from the generator stem using a sub-network and (ii) predicting future radar frames according to the self-supervised labels, enabling heterogeneous latent representation. Additionally, we attempt an ensemble forecast that prescribes random perturbations to improve performance. With the flexibility of representation learning, ClusterCast enables the model to learn precipitation distribution more accurately. Results indicate that our method generates non-blurry future frames by preventing mode collapse, and the proposed method demonstrates robustness across various precipitation scenarios. Extensive experiments demonstrate that our method outperforms four benchmarks on a 2-h prediction basis with a mean squared error (MSE) of 8.9% on unseen datasets.

Improved Error-Based Ensemble Learning Model for Compressor Performance Parameter Prediction

Large compressors have complex structures and constantly changing operating conditions. It is challenging to build physical models of compressors to analyse their performance parameters. An improved error-based stacked ensemble learning prediction model is proposed in this work. This model simplifies the modelling steps in a data-driven manner and obtains accurate prediction results. An enhanced integrated model employs K-fold cross-validation to assign dataset weights based on validation set errors, achieving a 12.4% reduction in average output error. Additionally, the output error of the meta-model undergoes a Box–Cox transformation for error compensation, decreasing the average output error by 14.0%. The Stacking model, combining the above improvements, notably reduces the root-mean-square errors for power, surge, and blocking boundaries by 24.2%, 20.6%, and 23.3%, respectively. This integration significantly boosts prediction accuracy.

In-silico prediction of dislodgeable foliar residues and regulatory implications for plant protection products.

When experimentally determined dislodgeable foliar residue (DFR) values are not available, regulatory agencies use conservative default DFR values as a first-tier approach to assess post-application dermal exposures to plant protection products (PPPs). These default values are based on a limited set of field studies, are very conservative, and potentially overestimate exposures from DFRs. Use Random Forest to develop classification and regression-type ensemble models to predict DFR values after last application (DFR0) by considering experimentally-based variability due to differences in physical and chemical properties of PPPs, agronomic practices, crop type, and climatic conditions. Random Forest algorithm was used to develop in-silico ensemble DFR0 prediction models using more than 100 DFR studies from Corteva AgriscienceTM. Several variables related to the active ingredient (a.i.) that was applied, crop, and climate conditions at the time of last application were considered as model parameters. The proposed ensemble models demonstrated 98% prediction accuracy that if a DFR0 is predicted to be less than the European Food Safety Authority (EFSA) default DFR0 value of 3 µg/cm2/kg a.i./ha, it is highly indicative that the measured DFR value will be less than the default if the study is conducted. If a value is predicted to be larger than or equal to the EFSA default, the model has an 83% prediction accuracy. This manuscript is expected to have significant impact globally as it provides: A framework for incorporating in silico DFR data into worker exposure assessment, A roadmap for a tiered approach for conducting re-entry exposure assessment, and A proof of concept for using existing DFR data to provide a read-across framework that can easily be harmonized across all regulatory agencies to provide more robust assessments for PPP exposures.

Diversity supporting robustness: Enhancing adversarial robustness via differentiated ensemble predictions

Deep learning models remain susceptible to adversarial attacks, prompting a primary focus on individual defense strategies. Previous studies have shown that increasing diversity among ensemble models can enhance their adversarial robustness. However, existing diversity metrics often lack a direct association with robustness, potentially limiting ensemble methods’ effectiveness in countering adversarial attacks. This paper presents a new diversity training method, termed Enhancing Adversarial Robustness through Diversity that Supports Robustness (EADSR), which is strongly linked to increased adversarial robustness. We developed a safety space hypothesis and a simultaneous training method based on the models’ responsiveness to non-natural perturbations which amplifies ensemble model diversity. By categorizing the ensemble models’ prediction behaviors into four classes, four regularization methods are applied for diversity training throughout the training process. We conducted comparative analysis with existing diversity methods to find a remarkable 30%–100% improvement on multiple benchmark datasets. The EADSR method incurs manageable training costs and is applicable to complex datasets such as ImageNet, demonstrating its effectiveness and practicality as a defense approach.

Uncertainty propagation analysis for distributed hydrological forecasting using a neural network

AbstractThe last few decades have presented a significant increase in hydrological disasters, such as floods. In some countries, most of the environmental, socioeconomic, and biodiversity losses are caused by floods. Thus, flood forecasting is crucial to support an efficient disaster warning system. This work proposes a model for hydrological forecasting based on a neural network with a geographically aligned input named GeoNN. It employs weather radar data to obtain accumulated rainfall in each grid cell of the watershed and make 15‐ and 120‐min predictions of the outlet river level. An uncertainty propagation analysis was performed for GeoNN from a collection of test cases obtained by either using different schemes of the dataset partitioning or introducing different additive‐noise rates to the input data to provide a probability of flood occurrence and also an ensemble prediction. Both this probability and the ensemble were able to detect occurrences of river levels exceeding a given flood threshold.

Ensemble learning prediction framework for EGFR amplification status of glioma based on terahertz spectral features

Epidermal growth factor receptor (EGFR) plays a pivotal role in the initiation and progression of gliomas. In particular, in glioblastoma, EGFR amplification emerges as a catalyst for invasion, proliferation, and resistance to radiotherapy and chemotherapy. Current approaches are not capable of providing rapid diagnostic results of molecular pathology. In this study, we propose a terahertz spectroscopic approach for predicting the EGFR amplification status of gliomas for the first time. A machine learning model was constructed using the terahertz response of the measured glioma tissues, including the absorption coefficient, refractive index, and dielectric loss tangent. The novelty of our model is the integration of three classical base classifiers, i.e., support vector machine, random forest, and extreme gradient boosting. The ensemble learning method combines the advantages of various base classifiers, this model has more generalization ability. The effectiveness of the proposed method was validated by applying an individual test set. The optimal performance of the integrated algorithm was verified with an area under the curve (AUC) maximum of 85.8 %. This signifies a significant stride toward more effective and rapid diagnostic tools for guiding postoperative therapy in gliomas.

Economics of physics-based solar forecasting in power system day-ahead scheduling

A high-quality solar power forecasting system that strictly adheres to grid regulations is valuable for system operators to formulate strategies for power system scheduling. Some grid operators, therefore, enact penalty schemes for the forecasts submitted by photovoltaic (PV) plant owners, as a means to fortify truthful and high-quality forecast submissions. From the perspectives of both plant owners and grid operators, this study inquires into the quality-to-value mapping of solar forecasts in the context of power system day-ahead scheduling. A physics-based solar power forecasting method is presented, which consists of two steps. Firstly, ensemble numerical weather prediction (NWP) is summarized into point forecasts. Then irradiance is converted to power via a physical model chain. The results reveal that the two-step physics-based forecasting method has an advantage over a winning method in Global Energy Forecasting Competition 2014 in terms of several accuracy measures. Subsequently, the economics of solar forecasting is quantified through performing day-ahead scheduling on a modified IEEE 30-bus system with PV and battery storage. It is demonstrated that, by respecting the statistical theory on consistency and elicitability when extracting point forecasts from NWP ensembles, both power system operators and PV plant owners can benefit profoundly in terms of cost savings. The former sees fewer needs for reserves, while the latter is less penalized. The data and Python code used to produce the results are also provided to enhance the reproducibility of this work.

A Bayesian Approach for Forecasting the Probability of Large Earthquakes Using Thermal Anomalies from Satellite Observations

Studies have demonstrated the potential of satellite thermal infrared observations to detect anomalous signals preceding large earthquakes. However, the lack of well-defined precursory characteristics and inherent complexity and stochasticity of the seismicity continue to impede robust earthquake forecasts. This study investigates the potential of pre-seismic thermal anomalies, derived from five satellite-based geophysical parameters, i.e., skin temperature, air temperature, total integrated column water vapor burden, outgoing longwave radiation (OLR), and clear-sky OLR, as valuable indicators for global earthquake forecasts. We employed a spatially self-adaptive multiparametric anomaly identification scheme to refine these anomalies, and then estimated the posterior probability of an earthquake occurrence given observed anomalies within a Bayesian framework. Our findings reveal a promising link between thermal signatures and global seismicity, with elevated forecast probabilities exceeding 0.1 and significant probability gains in some strong earthquake-prone regions. A time series analysis indicates probability stabilization after approximately six years. While no single parameter consistently dominates, each contributes precursory information, suggesting a promising avenue for a multi-parametric approach. Furthermore, novel anomaly indices incorporating probabilistic information significantly reduce false alarms and improve anomaly recognition. Despite remaining challenges in developing dynamic short-term probabilities, rigorously testing detection algorithms, and improving ensemble forecast strategies, this study provides compelling evidence for the potential of thermal anomalies to play a key role in global earthquake forecasts. The ability to reliably estimate earthquake forecast probabilities, given the ever-present threat of destructive earthquakes, holds considerable societal and ecological importance for mitigating earthquake risk and improving preparedness strategies.

Enhancing Weighted Averaging for CNN Model Ensemble in Plant Diseases Image Classification

Deep learning, especially convolutional neural networks (CNN), has gained traction in the field of image classification. In the specific case of plant disease classification, improving the accuracy and reliability of image classification is paramount. This paper delves into the ensemble prediction technique using a weighted soft-voting method. Instead of assigning a generalized weight to each CNN model, our approach emphasizes giving weights to each label's prediction within every individual model. We employed three esteemed CNN architectures for our experiments: DenseNet201, InceptionV3, and Xception focusing on classifying various diseases affecting grapes. By harnessing transfer learning coupled with end-to-end fine-tuning, we achieved a streamlined and efficient training process. Notably, the f1-score for each grape disease class was used as a parameter for weight determination and as a metric for the final evaluation. In our study, the newly proposed method was tested across various datasets and ensemble scenarios, demonstrating its effectiveness by not only outperforming the conventional soft-voting and prevalent weighted soft-voting methods, which achieved best scores of 95.68% and 95.81% respectively, but also by achieving a remarkable accuracy of 96.56%. This method's efficacy is heightened when ensemble models exhibit distinct characteristics; the more varied the model characteristics, the more enhanced the ensemble results.

Signature-based Adaptive Cloud Resource Usage Prediction Using Machine Learning and Anomaly Detection

One of the challenges in managing cloud computing clusters is assigning resources based on the customers’ needs. For this mechanism to work efficiently, it is imperative that there are sufficient resources reserved to maintain continuous operation, but not too much to avoid overhead costs. Additionally, to avoid the overhead of acquisition time, it is important to reserve resources sufficiently in advance. This paper presents a novel reliable general-purpose mechanism for prediction-based resource usage reservation. The proposed solution should be capable of operating for long periods of time without drift-related problems, and dynamically adapt to changes in system usage. To achieve this, a novel signature-based ensemble prediction method is presented, which utilizes multiple distinct prediction algorithms suited for various use-cases, as well as an anomaly detection mechanism used to improve prediction accuracy. This ensures that the mechanism can operate efficiently in different real-life scenarios. Thanks to a novel signature-based selection algorithm, it is possible to use the best available prediction algorithm for each use-case, even over long periods of time, which would typically lead to drifts. The proposed approach has been evaluated using real-life historical data from various production servers, which include traces from more than 1,500 machines collected over more than a year. Experimental results have demonstrated an increase in prediction accuracy of up to 21.4 percent over the neural network approach. The evaluation of the proposed approach highlights the importance of choosing the appropriate prediction method, especially in diverse scenarios where the load changes frequently.

A big data-based ensemble for fault prediction in electrical secondary distribution network

A big data-based ensemble for fault prediction in electrical secondary distribution network

Enhancing Crop Yield Predictions with PEnsemble 4: IoT and ML-Driven for Precision Agriculture

This research introduces the PEnsemble 4 model, a weighted ensemble prediction model that integrates multiple individual machine learning models to achieve accurate maize yield forecasting. The model incorporates unmanned aerial vehicle (UAV) imagery and Internet of Things (IoT)-based environmental data, providing a comprehensive and data-driven approach to yield prediction in maize cultivation. Considering the projected growth in global maize demand and the vulnerability of maize crops to weather conditions, improved prediction capabilities are of paramount importance. The PEnsemble 4 model addresses this need by leveraging comprehensive datasets encompassing soil attributes, nutrient composition, weather conditions, and UAV-captured vegetation imagery. By employing a combination of Huber and M estimates, the model effectively analyzes temporal patterns in vegetation indices, in particular CIre and NDRE, which serve as reliable indicators of canopy density and plant height. Notably, the PEnsemble 4 model demonstrates a remarkable accuracy rate of 91%. It advances the timeline for yield prediction from the conventional reproductive stage (R6) to the blister stage (R2), enabling earlier estimation and enhancing decision-making processes in farming operations. Moreover, the model extends its benefits beyond yield prediction, facilitating the detection of water and crop stress, as well as disease monitoring in broader agricultural contexts. By synergistically integrating IoT and machine learning technologies, the PEnsemble 4 model presents a novel and promising solution for maize yield prediction. Its application holds the potential to revolutionize crop management and protection, contributing to efficient and sustainable farming practices.

Brain Damage Detection Using Ensemble Classifier for Children Health Progress

Magnetic Resonance Imaging (MRI) is a medical imaging technique used to detect brain damage, particularly in children. Previous classification techniques for brain damage had low accuracy, complex edge detection, and high noise levels. In this proposed work, use an ensemble classification technique to analyze MRI brain images of 6-10 year old children. The input data is preprocessed using a Median filter to remove noise and improve image quality before being sent to a segmentation technique. Each part of the segmented image, such as tissue, inner layer, overall shape, and texture, is classified based on pixel size. The classification is done using an ensemble of Support Vector Machines (SVM) and K-Nearest Neighbors (KNN) techniques. The ensemble prediction is multiplied with several input values to classify the MRI image, aggregating different layer values. The output classified image is validated and developed using Mat Lab software simulation, with an accuracy of 97.86%. This method can accurately classify various types of brain damage and aid in early detection in children.

Ensemble tree model prediction of summer precipitation in North China based on predictor selection strategy

Ensemble tree model prediction of summer precipitation in North China based on predictor selection strategy

Ensemble machine learning prediction of hyperuricemia based on a prospective health checkup population.

Objectives: An accurate prediction model for hyperuricemia (HUA) in adults remain unavailable. This study aimed to develop a stacking ensemble prediction model for HUA to identify high-risk groups and explore risk factors. Methods: A prospective health checkup cohort of 40899 subjects was examined and randomly divided into the training and validation sets with the ratio of 7:3. LASSO regression was employed to screen out important features and then the ROSE sampling was used to handle the imbalanced classes. An ensemble model using stacking strategy was constructed based on three individual models, including support vector machine, decision tree C5.0, and eXtreme gradient boosting. Model validations were conducted using the area under the receiver operating characteristic curve (AUC) and the calibration curve, as well as metrics including accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and F1 score. A model agnostic instance level variable attributions technique (iBreakdown) was used to illustrate the black-box nature of our ensemble model, and to identify contributing risk factors. Results: Fifteen important features were screened out of 23 clinical variables. Our stacking ensemble model with an AUC of 0.854, outperformed the other three models, support vector machine, decision tree C5.0, and eXtreme gradient boosting with AUCs of 0.848, 0.851 and 0.849 respectively. Calibration accuracy as well as other metrics including accuracy, specificity, negative predictive value, and F1 score were also proved our ensemble model's superiority. The contributing risk factors were estimated using six randomly selected subjects, which showed that being female and relatively younger, together with having higher baseline uric acid, body mass index, γ-glutamyl transpeptidase, total protein, triglycerides, creatinine, and fasting blood glucose can increase the risk of HUA. To further validate our model's applicability in the health checkup population, we used another cohort of 8559 subjects that also showed our ensemble prediction model had favorable performances with an AUC of 0.846. Conclusion: In this study, the stacking ensemble prediction model for HUA was developed, and it outperformed three individual models that compose it (support vector machine, decision tree C5.0, and eXtreme gradient boosting). The contributing risk factors were identified with insightful ideas.

Применение стохастических возмущений параметров в версии модели ПЛАВ для долгосрочных прогнозов

The paper examines the impact of applying the disturbances of parameters for the schemes describing subgrid-scale processes in the SL-AV atmosphere model on the characteristics of average monthly atmospheric circulation in ensemble subseasonal forecasts. 25-year retrospective forecasts are computed for different seasons with different perturbation sets, the results are compared with forecasts without perturbations. It is shown that perturbing three parameters reduces forecast errors of some variables as compared to undisturbed forecasts without introducing errors into integral characteristics