Support Vector Regression Approach Research Articles

Forecasting Chinese crude oil futures volatility: New evidence based on dual feature processing of large‐scale variables

AbstractThis paper aims to study the volatility forecasting of Chinese crude oil futures from the large‐scale variables perspective by considering both the information on international futures markets volatility and technical indicators of Chinese crude oil futures. We employ the dual feature processing method (LASSO‐PCA) by integrating least absolute shrinkage and selection operator (LASSO) and principal component analysis (PCA) to extract important factors of the large‐scale exogenous variables. Besides the traditional ordinary least squares (OLS) estimation, the nonlinear support vector regression (SVR) approach is employed to integrate with the LASSO‐PCA method. The empirical results show that both the OLS and SVR combined with LASSO‐PCA can improve the forecasting accuracy, especially SVR‐LASSO‐PCA owns the best forecasting performance. The analysis of the selected variables finds international futures volatility is chosen more frequently. These results are further validated through a series of robust experiments. Finally, the time difference between China and the United States is also considered in order to obtain more reasonable out‐of‐sample forecasting.

Nondestructive detection of SSC in multiple pear (Pyrus pyrifolia Nakai) cultivars using Vis-NIR spectroscopy coupled with the Grad-CAM method

Vis-NIR spectroscopy coupled with chemometric models is frequently used for pear soluble solid content (SSC) prediction. However, the model robustness is challenged by the variations in pear cultivars. This study explored the feasibility of developing universal models for predicting SSC of multiple pear varieties to improve the model's generalizability. The mature fruits of 6 pear cultivars with green skin (Pyrus pyrifolia Nakai cv. ‘Cuiyu’, ‘Sucui No.1’ and ‘Cuiguan’) and brown skin (Pyrus pyrifolia Nakai cv. ‘Hosui’,’Syusui’ and ‘Wakahikari’) were used to establish single-cultivar models and multi-cultivar universal models using convolutional neural network (CNN), partial least square (PLS), and support vector regression (SVR) approaches. Multi-cultivar universal models were built using full spectra and important variables extracted by gradient-weighted class activation mapping (Grad-CAM), respectively. The universal models based on important variables obtained satisfactory performances with RMSEPs of 0.76, 0.59, 0.80, 1.64, 0.98, and 1.03°Brix on 6 cultivars, respectively.

Benchmarking machine learning and parametric methods for genomic prediction of feed efficiency-related traits in Nellore cattle

Genomic selection (GS) offers a promising opportunity for selecting more efficient animals to use consumed energy for maintenance and growth functions, impacting profitability and environmental sustainability. Here, we compared the prediction accuracy of multi-layer neural network (MLNN) and support vector regression (SVR) against single-trait (STGBLUP), multi-trait genomic best linear unbiased prediction (MTGBLUP), and Bayesian regression (BayesA, BayesB, BayesC, BRR, and BLasso) for feed efficiency (FE) traits. FE-related traits were measured in 1156 Nellore cattle from an experimental breeding program genotyped for ~ 300 K markers after quality control. Prediction accuracy (Acc) was evaluated using a forward validation splitting the dataset based on birth year, considering the phenotypes adjusted for the fixed effects and covariates as pseudo-phenotypes. The MLNN and SVR approaches were trained by randomly splitting the training population into fivefold to select the best hyperparameters. The results show that the machine learning methods (MLNN and SVR) and MTGBLUP outperformed STGBLUP and the Bayesian regression approaches, increasing the Acc by approximately 8.9%, 14.6%, and 13.7% using MLNN, SVR, and MTGBLUP, respectively. Acc for SVR and MTGBLUP were slightly different, ranging from 0.62 to 0.69 and 0.62 to 0.68, respectively, with empirically unbiased for both models (0.97 and 1.09). Our results indicated that SVR and MTGBLUBP approaches were more accurate in predicting FE-related traits than Bayesian regression and STGBLUP and seemed competitive for GS of complex phenotypes with various degrees of inheritance.

An efficient 2-D flood inundation modelling based on a data-driven approach

Study regionThe plain areas of Yilan County in Taiwan. Study focusThe present study employed hourly precipitation and inundation simulation data from a 5-km resolution HiRAM-WRF (High-Resolution Atmospheric Model-Weather Research and Forecasting) downscaled model and a 40-m resolution SOBEK model to construct a data-driven two-dimensional flood inundation model (D2DFIM) based on the support vector regression (SVR) approach. New hydrological insights for the regionThe dataset comprised 149 rainfall events and corresponding inundation events. We extracted the inundated and interpolated rainfall grid points distributed across Yilan County, Taiwan, from the SOBEK and HiRAM-WRF models. The D2DFIM (data-driven two-dimensional flood inundation model) utilized the parallel computing approach in the training phase, saving 90% of the elapsed time. The comparisons between the data-driven and physical-based models indicate that the D2DFIM could perform better in flood simulation if the input features comprised previous inundation depth and the last three time steps of rainfall information. Regarding the minimal required number of training samples, we found almost no significant difference in the flooding simulation once the training samples equalled or exceeded 50. Additionally, the elapsed time of D2DFIM is 9–10 times faster than that of SOBEK. We validated the well-developed D2DFIM using three historical inundated events induced by typhoons in Yilan County; the flood extents derived from the D2DFIM agreed well with the survey results.

Improving the quantification of fine particulates (PM2.5) concentrations in Malaysia using simplified and computationally efficient models

Air pollution assessment in urban and rural areas is really challenging due to high spatio-temporal variability of aerosols and pollutants and the uncertainties in measurements and modelling estimates. Nevertheless, accurate determination of the pollution sources and distribution of PM2.5 concentrations is especially important for source apportionment and mitigation strategies. This study provides estimates of PM2.5 concentrations across Malaysia in high spatial resolution, based on multi-satellite data and machine learning (ML) models, namely Random Forest (RF), Support Vector Regression (SVR) and extreme Gradient Boosting (XGBoost), also covering remote areas without measurement networks. The study aims to develop ML models that are simpler than previous works and demonstrate computational efficiency. Six sub-models were developed to represent different locations and seasons in Malaysia. Model 1 includes all data from 65 air-quality stations, Models 2 and 3 characterize urban/industrial and suburban sites, respectively, while Models 4 to 6 correspond to dry, wet, and inter-monsoon seasons, respectively. The RF technique exhibited slightly better performance compared to the XGBoost and SVR approaches. More specifically, for model 1, it exhibited a high correlation with a coefficient of determination (R2) of 0.64 and RMSE of 12.17 μg m−3, while similar results were obtained for models 3, model 4 and model 5. The lower performance (R2 = 0.16–0.94) observed in the wet and inter-monsoon seasons is due to fewer number of data used in model calibration. Integration of two AOD products from the Advanced Himawari Imager (AHI) and Visible Infrared Imaging Radiometer Suite (VIIRS) sensors together with gases pollutants from Sentinel 5P enabled seamless seasonal PM2.5 mapping over Malaysia, even for a short period of time. However, usage of data with insufficient information during the model training procedure, and lack of satellite data due to cloud contamination, can limit the PM2.5 prediction accuracy.

Systemic risk prediction using machine learning: Does network connectedness help prediction?

The global financial crisis not only highlights the important role of financial network connectedness, but also urges regulators pay more attention to systemic risk. We study whether network connectedness helps systemic risk prediction by using machine learning techniques, including k-nearest neighbor (KNN), support vector regression (SVR), extreme gradient boosting (XGB), and feedforward neural network (FNN). Based on prediction results, we use the fingerprint model to deliver interpretability evaluations of financial variables and network connectedness. We find that (1) the linear and nonlinear effects of stock market volatility and network connectedness effectively interpret systemic risk; (2) the interaction effects between stock market volatility and other drivers have a stronger ability to predict system risk in SVR, KNN, and XGB approaches, while network connectedness performs more profoundly in FNN approach; and (3) the interaction effect between stock market volatility and network connectedness is evident in all approaches.

Data-driven autoencoder neural network for onboard BMS Lithium-ion battery degradation prediction

An autoencoder based neural network architecture, CD-Net, is proposed to predict Lithium-ion battery capacity degradation as a function of operation time as part of a battery management system. CD-Net's generalization performance on various LIB cell chemistry is tested. The incorporation of cell chemistry in CD-Net leads to an improvement in the overall battery capacity prediction accuracy of >2 % for LiNiMnCoO2 cells, >5 % for LiNiCoAlO2 cells, and >12 % for LiFePO4 cells when compared to the similar ML models that do not incorporate cell chemistry information. A comparison of onboard battery health prediction using CD-Net against support vector regression, Bayesian regression, and Gaussian process regression-based approaches shows that CD-Net has higher computational efficiency with <2 % of relative remaining useful life (RUL) prediction error in a no-cell chemistry information setting. In summary, our work presents a chemistry-independent neural network model tailored specifically for onboard BMS applications, showcasing notable predictive capabilities in the context of Lithium-ion battery health assessment.

Assessment of the Support Vector Regression and Random Forest Algorithms in the Bias Correction Process on Temperatures

Climate information can be obtained from General circulation models (GCMs). However, this model has poor resolution, so it is necessary to do bias correction to overcome this problem. This study carried out a bias correction process using the Support Vector Regression (SVR) and Random Forest (RF) approaches. Bias correction is carried out for temperature in Indonesia using the BNU-ESM and MERRA-2 climate models, which act as observational data. The results show that the RF method (RMSE: 0.334; Correlation: 0.694; Standard Deviation: 0.582) is better than SVR (RMSE: 0.341; Correlation: 0.675; Standard Deviation: 0.588) in performing bias correction.

Support Vector Regression for Bluetooth Ranging in Multipath Environments

Ranging solutions for Internet of Things (IoT) localization applications seek to provide high accuracy with low cost of implementation. Among candidate IoT technologies that may fit this criterion, Bluetooth is a desirable choice as Bluetooth Low Energy (BLE) support is ubiquitous in modern smartphones, providing low implementation costs and low power consumption. Recent advancements in BLE ranging technology employ the Multi-Carrier Phase Difference technique which takes two-way channel frequency response (CFR) measurements. However, accurate ranging with these measurements is challenging due to many closely spaced multipath components from squaring the one-way CFR, a single or low number of snapshots, and model imperfections that arise in practical scenarios. To overcome these challenges, we propose a data-driven support vector regression (SVR) approach. Using real-world BLE measurements, our proposed SVR method demonstrates decimeter-level accuracy with single antenna devices, whereas Multiple Signal Classification (MUSIC), a popular model-based method, requires multiple antennas to obtain comparable performance. Moreover, we show robustness in different multipath environments including indoor, outdoor, and non-line-of-sight conditions, we determine generalization capabilities with training size, and we analytically establish the reduction in computational complexity compared to MUSIC.

Strategic combination of waste plastic/tire pyrolysis oil with biodiesel for natural gas-enriched HCCI engine: Experimental analysis and machine learning model

In this experiment, different combinations and blends based on 50% biodiesel and 50% pyrolysis oil were prepared to create 4 fuel samples for all tests. These samples were provided to the test engine operated on the conventional mode and homogeneous charge compression-ignition (HCCI) mode aiming to evaluate the performance, emission, and combustion characteristics of these modes. In the HCCI mode, a steady flow of 3 L per minute of compressed natural gas (CNG) was injected together with the air. As a result, preheated 50% Pongamia biodiesel/50% plastic pyrolysis oil combined with enriched CNG for the HCCI mode was found to be superior to those of other fuels according to performance, combustion, and emission characteristics although brake thermal efficiency was slightly lower than the conventional diesel engine. In addition, the performance and emission parameters of the HCCI engine were also predicted by using three machine learning models such as Decision Tree, Random Forest, and Support Vector Regression. Finally, Random Forest and Support Vector Regression approaches were recommended to predict the operation parameters of the HCCI engine with an accuracy level as high as nearly 99% in most instances, except for HC and CO with 80% and 90% of accuracy levels, respectively.

Generalized Support Vector Regression and Symmetry Functional Regression Approaches to Model the High-Dimensional Data

The analysis of the high-dimensional dataset when the number of explanatory variables is greater than the observations using classical regression approaches is not applicable and the results may be misleading. In this research, we proposed to analyze such data by introducing modern and up-to-date techniques such as support vector regression, symmetry functional regression, ridge, and lasso regression methods. In this study, we developed the support vector regression approach called generalized support vector regression to provide more efficient shrinkage estimation and variable selection in high-dimensional datasets. The generalized support vector regression can improve the performance of the support vector regression by employing an accurate algorithm for obtaining the optimum value of the penalty parameter using a cross-validation score, which is an asymptotically unbiased feasible estimator of the risk function. In this regard, using the proposed methods to analyze two real high-dimensional datasets (yeast gene data and riboflavin data) and a simulated dataset, the most efficient model is determined based on three criteria (correlation squared, mean squared error, and mean absolute error percentage deviation) according to the type of datasets. On the basis of the above criteria, the efficiency of the proposed estimators is evaluated.

Support vector regression approach to optimize the biodiesel composition for improved engine performance and lower exhaust emissions

The practical applicability of biodiesel is limited due to poor fuel economy and higher emissions of oxides of nitrogen. Moreover, engine characteristics vary significantly with biodiesel type, highlighting the necessity to determine the most suitable biodiesel for improved fuel economy and emissions. Choosing the most suitable biodiesel is challenging as the relationship between biodiesel composition and engine characteristics is complex. Thus, it is intended to determine the most suitable biodiesel by developing composition-based models that predict engine characteristics and use them for optimization. This study explores the applicability of support vector machine regression to correlate biodiesel composition with engine characteristics. Two additional biodiesels not used for calibration validate the developed models with an acceptable mean absolute percentage error of less than 7%. Further, the models are fed to the genetic algorithm to determine the optimal composition for lower brake specific fuel consumption (BSFC) and emissions of unburnt hydrocarbon (HC), carbon monoxide (CO), and oxides of nitrogen (NOx). The optimized biodiesel has 8% shorter chain esters and 92% longer chain esters. Sunflower biodiesel exhibits 23% higher BSFC than diesel, whereas for optimized biodiesel, it is 4% higher. The proposed biodiesel results in 42% and 2% less HC and CO emissions than diesel and are reasonably lower than other biodiesels. The optimized biodiesel results in 42% higher NOx emissions, while sunflower biodiesel exhibits 83% higher NOx emissions than diesel. Overall, the optimized biodiesel resulted in a minimum penalty in fuel economy and NOx emissions compared to other commercially available biodiesels.

Vertical and Horizontal Water Penetration Velocity Modeling in Nonhomogenous Soil Using Fast Multi-Output Relevance Vector Regression.

A joint determination of horizontal and vertical movement of water through porous medium is addressed in this study through fast multi-output relevance vector regression (FMRVR). To do this, an experimental data set conducted in a sand box with 300 × 300 × 150 mm dimensions made of Plexiglas is used. A random mixture of sand having size of 0.5-1 mm is used to simulate the porous medium. Within the experiments, 2, 3, 7, and 12 cm walls are used together with different injection locations as 130.7, 91.3, and 51.8 mm measured from the cutoff wall at the upstream. Then, the Cartesian coordinated of the tracer, time interval, length of the wall in each setup, and two dummy variables for determination of the initial point are considered as independent variables for joint estimation of horizontal and vertical velocity of water movement in the porous medium. Alternatively, the multi-linear regression, random forest, and the support vector regression approaches are used to alternate the results obtained by the FMRVR method. It was concluded that the FMRVR outperforms the other models, while the uncertainty in estimation of horizontal penetration is larger than the vertical one.

Ellipsoidal classification via semidefinite programming

We propose a classification approach exploiting relationships between ellipsoidal separation and Support-vector Machine (SVM) with quadratic kernel. By adding a (Semidefinite Programming) SDP constraint to SVM model we ensure that the chosen hyperplane in feature space represents a non-degenerate ellipsoid in input space. This allows us to exploit SDP techniques within Support-vector Regression (SVR) approaches, yielding better results in case ellipsoid-shaped separators are appropriate for classification tasks. We compare our approach with spherical separation and SVM on some classification problems.

Spatial Prediction of Wildfire Susceptibility Using Hybrid Machine Learning Models Based on Support Vector Regression in Sydney, Australia

Australia has suffered devastating wildfires recently, and is predisposed to them due to several factors, including topography, meteorology, vegetation, and ignition sources. This study utilized a geographic information system (GIS) technique to analyze and understand the factors that regulate the spatial distribution of wildfire incidents and machine learning to predict wildfire susceptibility in Sydney. Wildfire inventory data were constructed by combining the fire perimeter through field surveys and fire occurrence data gathered from the visible infrared imaging radiometer suite (VIIRS)-Suomi thermal anomalies product between 2011 and 2020 for the Sydney area. Sixteen wildfire-related factors were acquired to assess the potential of machine learning based on support vector regression (SVR) and various metaheuristic approaches (GWO and PSO) for wildfire susceptibility mapping in Sydney. In addition, the 2019–2020 “Black Summer” fire acted as a validation dataset to assess the predictive capability of the developed model. Furthermore, the information gain ratio (IGR) method showed that driving factors such as land use, forest type, and slope degree have a large impact on wildfire susceptibility in the study area, and the frequency ratio (FR) method represented how the factors influence wildfire occurrence. Model evaluation based on area under the curve (AUC) and root average square error (RMSE) were used, and the outputs showed that the hybrid-based SVR-PSO (AUC = 0.882, RMSE = 0.006) model performed better than the standalone SVR (AUC = 0.837, RMSE = 0.097) and SVR-GWO (AUC = 0.873, RMSE = 0.080) models. Thus, optimizing SVR with metaheuristics improved the accuracy of wildfire susceptibility modeling in the study area. The proposed framework can be an alternative to the modeling approach and can be adapted for any research related to the susceptibility of different disturbances.

Longitudinal Effects of 1-Year Smoking Cessation on Human Bronchial Epithelial Transcriptome

Longitudinal Effects of 1-Year Smoking Cessation on Human Bronchial Epithelial Transcriptome

Support Vector Regression approach for prediction of delamination at entry and exit during drilling of GFRP Composites

The demand for Composites in the modern era have increased immensely due to its vast applications and superior properties over conventional materials. Glass Fibre Reinforced Plastic (GFRP) is one of the economic alternative to conventional engineering materials due to its high specific modulus of elasticity, high specific strength, good corrosion resistance, high fatigue strength and lightweight. Components made out from GFRP composites are usually near net shaped and require holes for assembly integration. Drilling is an important process as concentrated forces can cause major damage to the composite. Drilling of GFRP causes various damage such as thermal degradation, fibre breakage, matrix cracking and delamination. A substantial damage is caused by delamination which can occur both on the entry and exit sides of the composite, exit side delamination considered more severe. Therefore, selection of proper process parameters during drilling operation is very much essential. In the present work, a support vector regression (SVR) model is developed to predict the delamination at entry and exit during the drilling of GFRP composites. The model is developed based on the data obtained from experimentation. The model accuracy is evaluated by the three performance criteria including root mean square error (RMSE), Nash–Sutcliffe efficiency co-efficient (E) and co-efficient of determination (R2). The model provides an inexpensive and time saving alternative to study the delamination at entry and exit of the GFRP composite actual drilling operation.

Monthly streamflow prediction using hybrid extreme learning machine optimized by bat algorithm: a case study of Cheliff watershed, Algeria

ABSTRACT In the present paper, we propose a new approach for monthly streamflow prediction based on the extreme learning machine (ELM) and the metaheuristic bat algorithm (Bat-ELM). The performance of the Bat-ELM was compared to that of ELM, support vector regression (SVR), Gaussian process regression (GPR), multilayer perceptron neural network (MLPNN), and generalized regression neural network (GRNN). The proposed models were applied using data from three hydrometric stations located in the Cheliff Basin, Algeria. The results showed that the Bat-ELM was more satisfactory than the standalone models. The Bat-ELM achieved the highest numerical performance with correlation coefficient and Nash-Sutcliffe efficiency ranging from 0.927 to 0.973 and from 0.846 to 0.944, respectively, much higher than the respective values obtained using the MLPNN, GRNN, SVR, GPR and ELM approaches. The obtained results demonstrate that the Bat-ELM is an interesting alternative algorithm for predicting high and extreme streamflow.

Statistical analysis and prediction of force and overtopping rates on large-scale vertical walls using support vector machine and random forest regression

This study provides a statistical basis to determine the most influencing parameters on forces and overtopping over vertical walls, as well as to showcase the usability of machine learning modelling in coastal engineering. To this end, horizontal force and overtopping data for regular waves of varying height (0.63–1.65 m), period (4–8 s), and water depth (3.37–3.97 m) over a vertical wall were studied using redundancy analysis (RDA) and regressed using multiple linear regression, support vector regression (SVR), and random forest regression (RFR). The RDA showed that about 60% of the output variable variance can be explained by the structure dimensions and 15% by the incoming wave characteristics. The SVR approach better predicted the average force (mean relative error (MRE) = 39.9% and R2 = 0.346), whereas the RFR technique better predicted overtopping discharges (MRE = 46.7% and R2 = 0.802). By expanding the database, the error on overtopping prediction was reduced to 22.1% and 27.5%, respectively, for the SVR and RFR.

Conceptual hydrological model-guided SVR approach for monthly lake level reconstruction in the Tibetan Plateau

Study regionTibetan Plateau (TP) Study focusLakes in the TP that are subject to low human activity serve as an important indicator for quantitative assessment of regional climate change. However, previous studies have mainly focused on annual changes in lake area, level, and storage because of limited monitoring stations, and long-term lake level reconstruction at monthly resolution remains challenging. We propose a conceptual hydrological model-guided monthly lake level reconstruction (WBM-SVR) approach that incorporates water balance models (WBMs) and support vector regression (SVR), to improve the training and testing sets compared with SVR alone through consideration of the hydrological process. The WBM-SVR approach integrates WBMs to select input factors, sub-process control equations to quantify the contributions of input factors, empirical parameters to characterize catchment uniqueness, and SVR for water level modelling. New hydrological insights for the regionPhysically guided WBM-SVR is more accurate than SVR in reconstructing monthly lake water levels. WBMs can quantify the hydrological process in the lake catchment area with efficient quasi-physical mechanisms and refine the input–output factors within lake hydrometeorology. The reconstructability, generalization capability, and transferability of WBM-SVR are validated for three different types of lakes (glacier-free inflow lake, glacier-free outflow lake, and glacier-fed inflow lake), and the reconstruction results indicate significant improvements in WBM-SVR compared with SVR. The WBM-SVR approach shows great promise for achieving monthly lake level reconstruction.