Variable Importance Measurement Research Articles

Feature Extraction and Classification of Canopy Gaps Using GLCM- and MLBP-Based Rotation-Invariant Feature Descriptors Derived from WorldView-3 Imagery

Accurate mapping of selective logging (SL) serves as the foundation for additional research on forest restoration and regeneration, species diversification and distribution, and ecosystem dynamics, among other applications. This study aimed to model canopy gaps created by illegal logging of Ocotea usambarensis in Mt. Kenya Forest Reserve (MKFR). A texture-spectral analysis approach was applied to exploit the potential of WorldView-3 (WV-3) multispectral imagery. First, texture properties were explored in the sub-band images using fused grey-level co-occurrence matrix (GLCM)- and local binary pattern (LBP)-based texture feature extraction. Second, the texture features were fused with colour using the multivariate local binary pattern (MLBP) model. The G-statistic and Euclidean distance similarity measures were applied to increase accuracy. The random forest (RF) and support vector machine (SVM) were used to identify and classify distinctive features in the texture and spectral domains of the WV-3 dataset. The variable importance measurement in RF ranked the relative influence of sets of variables in the classification models. Overall accuracy (OA) scores for the respective MLBP models were in the range of 80–95.1%. The respective user’s accuracy (UA) and producer’s accuracy (PA) for the univariate LBP and MLBP models were in the range of 67–75% and 77–100%, respectively.

Quantitative Analysis of the Cu Element Enhanced by AgNPs in a Single Microsized Suspended Particle Based on Optical Trapping-LIBS and Machine Learning.

Extremely severe and persistent particulate pollution caused by industrialization and urbanization impacts air quality, regional and global climates, and human health. The unstable and complex spectral signal of laser-induced breakdown spectroscopy (LIBS) with minimal feature information and interference signals considerably influences the accuracy of qualitative and quantitative analysis. In response to overcome this phenomenon, in this work, quantitative analysis of Cu element enhanced by silver nanoparticles (AgNPs) in a single microsized suspended particle was proposed herein using optical trapping-LIBS and machine learning method was proposed. Initially, the optimal AgNPs enhancement conditions were optimized. The LIBS spectra of 15 polluted black carbon samples were collected and various spectral pretreatment methods were compared to optimize the LIBS spectra. Variable selection methods include variable importance measurement (VIM), variable importance projection (VIP), VIM-successive projections algorithm (VIM-SPA), VIM-genetic algorithm (VIM-GA), and VIM-mutual information (VIM-MI). Finally, several hybrid variable selection methods were implemented in random forest (RF) calibration models. In particular, a wavelet transform (WT)-VIM-SPA-RF calibration model has constructed under the WT spectral pretreatment method and the selected and optimized input variables (VIM-SPA). Results elucidate that the WT-VIM-SPA-RF calibration model (R2P = 0.9858, MREP = 0.0396) have the best prediction performance than the WT-RF and Raw-RF models in predicting the Cu level in a single microsized black carbon particle. Compared to the WT-RF and Raw-RF models, MREP values decreased by 37% and 62%, respectively. The values of RSD, RPD, and RER of this calibration model are 2.8%, 8.39%, and 17.79%, respectively. The aforementioned results demonstrate that the WT-VIM-SPA-RF calibration model with accuracy, stability, and robustness is a promising approach for improving the quantitative accuracy of the Cu level in carbon black particles.

Novel Random Forest Ensemble Modeling Strategy Combined with Quantitative Structure-Property Relationship for Density Prediction of Energetic Materials.

With the further development of the concept of green chemistry, the new generation of energetic materials tends to exhibit detonation properties such as higher insensitivity, higher density, and higher energy. Therefore, the precise molecular design and green and efficient synthesis of energetic materials will be one of the serious challenges. For the purpose of accurate prediction of detonation performance of energetic materials, an ensemble modeling strategy based on the combination of Monte Carlo (MC) and variable importance measurement (VIM) improved random forest (RF) and quantitative structure-property relationship (QSPR) is proposed, which was successfully used for density prediction of energetic materials. First, the structure of 162 energetic compounds was optimized by Gaussian software, and the molecular descriptor data were calculated by CODESSA software based on the optimized molecular structure. Then, the MCVIMRF_Med ensemble model was constructed on the basis of the above molecular descriptor data and the corresponding energetic compound density index. The joint X-Y distance algorithm (SPXY) is used to partition the data set. And then, MC is used to further divide the calibration set data into multiple subsets for the construction of the ensemble model. The subset size and the number of iterations of the MCVIMRF_Med ensemble model were optimized through MC cross validation. The final output strategy of the ensemble model is optimized based on the optimized parameters, and an output optimization method based on median screening is proposed and successfully applied for the prediction performance optimization of the MCVIMRF_Med ensemble model. To further investigate the performance of the MCVIMRF_Med ensemble model, the performance of it was compared with partial least squares, RF, VIMRF, and MCVIMRF calibration models. It shows that the MCVIMRF_Med ensemble model can achieve a better prediction result for the density of energetic materials, with R 2 CV of 0.9596, RMSECV of 0.0437 g/cm3, R 2 P of 0.9768, RMSEP of 0.0578 g/cm3, and relative analysis deviation of prediction set of 3.951. Therefore, the MCVIMRF_Med ensemble modeling strategy combined with QSPR is an effective approach for the density prediction of energetic materials. This work is expected to provide new research ideas and technical support for accurate prediction of detonation performance of energetic materials.

Multi-element Quantitative Analysis of Single Micro-sized Suspended Particles in Air with High Accuracy Based on Random Forest and Variable Selection Strategies.

The chemical compositions of atmospheric particles have been studied for several decades, and the traditional techniques for particle analysis usually require time-consuming sample preparation. Within this study, simultaneous quantitative detection of multiple metallic species (Zn, Cu, and Ni) in single micro-sized suspended particles was investigated by combining random forest (RF) and variable selection strategies. Laser-induced breakdown spectra of 15 polluted black carbon samples were applied for establishing the RF model, and the movmean smoothing spectral pretreatment method and variable selection methods [variable importance measurement (VIM), genetic algorithm (GA), and variable importance projection (VIP)] were proposed. Finally, the optimized RF calibration model with the evaluation indicators of mean relative error (MRE), root-mean-square error (RMSE), and coefficient of determination (R2) was constructed based on the optimal input variables and model parameters. Compared with the univariate regression method, the VIP-RF (Zn) and VIM-RF (Cu and Ni) models showed a better correlation relationship (Rp2 = 0.9662 for Zn, Rp2 = 0.9596 for Cu, and Rp2 = 0.9548 for Ni). For Zn, Cu, and Ni, the values of RMSEP (RMSE of prediction) decreased by 116.44, 68.94, and 102.10 ppm, while the values of MREP (MRE of prediction) decreased by 67, 55, and 48%, respectively. The values of ratio of prediction to deviation (RPD) of VIP-RF (Zn), VIM-RF (Cu), and VIM-RF (Ni) models were 5.4, 5.0, and 4.7, respectively. The performance of this combined approach displays a notable accuracy improvement in the quantitative analysis of single particles, suggesting that it is a promising tool for real-time air particulate matter pollution monitoring and control in the future.

Evaluation of the potential ecological risk of metals in atmospherically deposited particulate matter via laser-induced breakdown spectroscopy combined with machine learning

With the acceleration of industrialization and urbanization, China faces increasingly serious urban air pollution, such as frequent haze. Heavy metals have attracted extensive attention because of their non-degradability, significant biotoxicity, and persistence. In this work, laser-induced breakdown spectroscopy (LIBS) combined with random forest (RF) was utilized to directly evaluate the potential ecological risk of metals. The LIBS spectra of 17 atmospherically deposited particulate matter samples were collected, and three metal elements (V, Cr, and Zn) were identified from the National Institute of Standards and Technology database. The influence of predictive ability with different spectral preprocessing methods on the RF calibration model was then investigated, and the input variable was selected by variable importance measurement to further enhance the accuracy of predictive ability. Under the optimized spectral preprocessing method, VI threshold, and model parameters, the mean relative error of prediction set (MREP) for potential ecological risk index analysis (V, Cr, and Zn) was calculated as 0.0206, 0.0529, and 0.2218, respectively, and the determination coefficient of prediction set ( R P 2 ) was 0.9220, 0.9458, and 0.9359, respectively. These values inferred the overall lack of significant contamination at the time of the study. In summary, a novel approach based on LIBS combined with RF provides a new idea and method to directly evaluate the potential ecological risk of air pollution. LIBS combined with RF was utilized to direct evaluation of the potential ecological risk of metal elements in atmospherically deposited particulate matter.

A machine learning approach to mapping canopy gaps in an indigenous tropical submontane forest using WorldView-3 multispectral satellite imagery

SummarySelective logging in tropical forests may lead to deforestation and forest degradation, so accurate mapping of it will assist in forest restoration, among other ecological applications. This study aimed to track canopy tree loss due to illegal logging of the important hardwood tree Ocotea usambarensis in a closed-canopy submontane tropical forest by evaluating the mapping potential of the very-high-resolution WorldView-3 multispectral dataset using random forest (RF) and support vector machine (SVM) with radial basis function kernel classifiers. The results show average overall accuracies of 92.3 ± 2.6% and 94.0 ± 2.1% for the RF and SVM models, respectively. Average kappa coefficients were 0.88 ± 0.03 for RF and 0.90 ± 0.02 for SVM. The user’s and producer’s accuracies for both classifiers were in the range of 84–100%. This study further indicates that vegetation indices derived from bands 5 and 6 helped detect canopy gaps in the study area. Both variable importance measurement in the RF algorithm and pairwise feature selection proved useful in identifying the most pertinent variables in the classification of canopy gaps. These findings could allow forest managers to improve methods of detecting canopy gaps at larger scales using remote sensing data and relatively little additional fieldwork.

Maternal Levels of Cytokines in Early Pregnancy and Risk of Autism Spectrum Disorders in Offspring

Previous studies indicate a role of immune disturbances during early development in the etiology of autism spectrum disorders (ASD). Any potential disturbances during fetal development are best addressed by prospective evaluation of maternal markers of inflammation. Previous studies have investigated maternal cytokines, a group of powerful effectors of the immune system, with inconsistent results. In this study, we aimed to clarify the relationship between maternal cytokines and ASD by evaluating levels of 17 cytokines in first trimester maternal serum samples, from 318 mothers to ASD-cases and 429 mothers to ASD-unaffected controls, nested within the register-based Stockholm Youth Cohort. Overall, we observed no consistent associations between levels of maternal cytokines and ASD. While we observed a number of individual associations, the patterns varied across the diagnostic sub-groups. Levels above the 90th percentile of IL-1β (OR = 2.31, 95% CI 1.16–4.60), IL-7 (OR = 2.28, 95% CI 1.20–4.33), IL-13 (OR = 2.42, 95% CI 1.29–4.55), and MCP-1 (OR = 2.09, 95% CI 1.03–4.24) were associated with increased odds of ASD with co-occurring intellectual disability (ID), whereas GMCSF (OR = 2.06, 95% CI 1.03–4.11) and TNF-α (OR = 2.31, 95% CI 1.18–4.50) were associated with increased odds of ASD with ADHD but none survived correction for multiple comparisons. Also, none of the measured maternal cytokines were associated with ASD without co-occurring ID or ADHD. Implementing a data-driven approach using machine learning (Random Forest's Variable Importance measurement), we found no evidence to suggest that adding these cytokines and other markers of maternal immunity, to register-based maternal factors (e.g., psychiatric history) improves prediction of ASD. In summary, we found no robust evidence of an association between maternal immune markers during early pregnancy and ASD.

Discrimination of Radix Astragali according to geographical regions by data fusion of laser induced breakdown spectroscopy (LIBS) and infrared spectroscopy (IR) combined with random forest (RF)

Data fusion of Laser-induced breakdown spectroscopy (LIBS) and infrared spectroscopy (IR) combined with random forest (RF) was proposed to identify Radix Astragali from different geographical regions. Firstly, the LIBS and IR spectra of 19 Radix Astragali samples were collected and analyzed. Then, an unsupervised discriminant model based on principal components analysis (PCA) with single LIBS and IR spectra was applied to identify Radix Astragali. The results indicated that three types of Radix Astragali samples can not be accurately distinguished using PCA. To distinguish Radix Astragali accurately, a supervised discriminant model based on RF was applied to identify Radix Astragali. The optimal RF discriminant models based on single LIBS or IR were obtained by selecting appropriate pretreatment methods. In low-level data fusion, the optimal LIBS and IR spectra were concatenated into a matrix to construct RF discriminant model. In mid-level data fusion, feature variables were extracted by variable importance (VI) measurement on basis of low-level fused data and the optimal variables were used as input variables to construct RF discriminant model. The results indicated that the predictive performance of RF model based on data fusion was better than the single LIBS or IR method. The predictive performance of RF model based on mid-level data fusion was best and the corresponding sensitivity, specificity and accuracy for the test set were 0.9667, 0.9833 and 0.9778. Additionally, the modeling time was 6.9 s. Summarily, RF model on basis of data fusion of LIBS and IR can provide a rapid and accurate discriminant method for the Radix Astragali from different regions.

Performance prediction of suspension freeze crystallization for the treatment of liquid hazardous wastes via machine learning methods

The experimental method for determining treatment efficiency of suspension freeze crystallization (SFC) on liquid hazardous wastes (LHWs) is accurate but complex, costly and time-consuming. In the present study, artificial neural works (ANN) and random forest (RF), two machine learning methods, were utilized to develop models that were capable of predicting the treatment performance of SFC based on 8 typical LHWs. The targeted solutes in the chosen LHWs were characterized by COD, TOC, TDS, sulfide, conductivity, etc. The models were induced and tested to predict solute removal efficiency in accordance with freezing conditions and solution characteristics of 328 pieces of data collected from previous publications. Although both models have comparable predictive power, RF model presented better prediction accuracy and power (R2 = 0.9811, RMSE = 0.0323) than ANN model (R2 = 0.9615, RMSE = 0.0481). At the same time, the RF models showed better generalization ability than ANN models regardless different LHWs. The variable importance measurement indicated that ice phase fraction was the most important factor for solute removal efficiency in SFC process. The accurate predictability of developed models could be used before actual experiment to predict the removal efficiency of SFC according to various independent variables, so as to significantly reduce experiment workload of searching for the optimum freezing conditions. The variable importance measurement could provide a right direction for adjust the higher treatment efficiency of SFC on LHWs in the real situation.

Predicting co-pyrolysis of coal and biomass using machine learning approaches

Coal and biomass co-thermochemical conversion has caught significant attentions, in which the co-pyrolysis is always the primary process. The traditional pyrolysis kinetic models are developed individually for coal and biomass, in which the synergistic effect wasn’t comprehensively considered. In the present study, we innovatively explored a new method to accurately model this process using machine learning approaches, specifically the random forest algorithm based on classification and regression trees and extremely trees. First, a co-pyrolysis database is constructed from experimental data in published literatures, then divided into several sub-sets for training, application, and optimization, respectively. The machine learning models are trained on the training data-set, tested on the test data-set, and applicated on the new data-set. The training and test results demonstrate both models are able to well predict the co-pyrolysis (R2 > 0.999), and the application results demonstrate models also perform well at outside data (R2 > 0.873), with model based on extremely trees performs better owing to its better accuracy, generalization and less overfitting. It also demonstrates the known of biomass pyrolysis will be better than known of coal pyrolysis. In addition, the suggestion of input feature groups is given through parametric study, and variable importance measurement are explored.

Hybrid variable selection strategy coupled with random forest (RF) for quantitative analysis of methanol in methanol-gasoline via Raman spectroscopy

With the trend of portable and miniaturization, Raman spectrometer requires more advanced analytical methods providing more rapid and accurate analysis performance for in-situ analysis. In this work, a hybrid variable selection method based on V-WSP and variable importance measurement (VIM) coupled with random forest (RF) was used to improve the quantitative analysis performance of portable laser Raman instruments for quantitative analysis of methanol content in methanol gasoline. First, five preprocessing methods were applied to reduce the infection information in the raw spectra, respectively. Based on the spectra data processed by multivariate scattering correction (MSC), V-WSP was employed to filter the infection or redundant information in Raman spectroscopy, and 579 variables were obtained when the correlation threshold is 0.9600. Then, the variables were further eliminated by VIM. Finally, 43 variables were obtained by the V-WSP-VIM method. In data processing, out of bag (OOB) error estimation and 10-flod cross validation (CV) were applied to optimize the parameters of preprocessing methods, V-WSP, VIM and RF model. The results fully demonstrated that compared with the RF model based on raw spectra, the RF model based on V-WSP-VIM method can achieve a better prediction performance for the quantitative analysis of methanol content in methanol-gasoline, with the coefficients of determination of cross-validation (R2CV) improving from 0.9100 to 0.9662, the root mean square error of cross-validation (RMSECV) reducing from 0.0572 to 0.0365%, the coefficients of determination of prediction set (R2P) improving from 0.9214 to 0.9407, the root mean square error of prediction set (RMSEP) reducing from 0.0420 to 0.0382%, the variables reducing from 1044 to 43 and the modeling time reducing from 72.94 to 6.41 s. The results indicates that V-WSP-VIM coupled with RF is an effective method to improve the performance of portable laser Raman spectrometer for quantitative analysis of methanol content in methanol gasoline.

Modelling of coal flotation responses based on operational conditions by random forest

Coal consumption is one of the critical factors in the economy of China. Flotation separation of coal from its inorganic part (ash) can reduce environmental problems of coal consumption and improve its combustion. This investigation used random forest (RF) as an advanced machine learning method to rank flotation operations by variable importance measurement and predict flotation responses based on operational parameters. Fifty flotation experiments were designed, and performed based on various flotation conditions and by different variables (collector dosage, frother dosage, air flowrate, pulp density, and impeller speed). Statistical assessments indicated that there is a significant negative correlation between yield and ash content. Experiments indicated that in the optimum conditions, yield and ash content would be 80 and 9%, respectively. Variable importance measurement by RF showed that frother has the highest effectiveness on yield. Outcomes of modelling released that RF can accurately be used for ranking flotation parameters, and generating models within complex systems in mineral processing. [Received: May 20, 2020; Accepted: July 19, 2020]

Modelling of coal flotation responses based on operational conditions by random forest

Coal consumption is one of the critical factors in the economy of China. Flotation separation of coal from its inorganic part (ash) can reduce environmental problems of coal consumption and improve its combustion. This investigation used random forest (RF) as an advanced machine learning method to rank flotation operations by variable importance measurement and predict flotation responses based on operational parameters. Fifty flotation experiments were designed, and performed based on various flotation conditions and by different variables (collector dosage, frother dosage, air flowrate, pulp density, and impeller speed). Statistical assessments indicated that there is a significant negative correlation between yield and ash content. Experiments indicated that in the optimum conditions, yield and ash content would be 80 and 9%, respectively. Variable importance measurement by RF showed that frother has the highest effectiveness on yield. Outcomes of modelling released that RF can accurately be used for ranking flotation parameters, and generating models within complex systems in mineral processing. [Received: May 20, 2020; Accepted: July 19, 2020]

Prediction of specific gravity of Afghan coal based on conventional coal properties by stepwise regression and random forest

ABSTRACT Coal can be considered as the main fuel for electricity generation in Afghanistan. However, there is a quite limited data available about the overall quality, distribution, and character of coals in Afghanistan. Specific gravity (S.G) of coal as a key factor can be used for the estimation of potential tonnage production and be a fundamental parameter for the selection of coal washery process method. However, there is no investigation which comprehensively explores relationships between S.G and coal properties. In this investigation, the potential of S.G prediction based on conventional properties for Afghan coal samples was explored by stepwise regression and random forest. Pearson correlation (r) and variable importance measurement (VIM) of random forest (RF) were applied to select the most effective variables among conventional parameters for the S.G prediction. Results of VIM indicated that ash and carbon content of coal samples had the highest importance for the S.G prediction. Stepwise regression and RF models were developed based on these two coal variables. Testing the generated models indicated that S.G of Afghan coals can quite accurately predict by these models (R2 > 0.90). Modeling outcomes showed that the highest S.G (S.G > 2) for Afghan coal occurred when ash was higher than 40% and carbon was lower than 30%.

Factors Influencing the Difficulty Level of the Subject: Machine Learning Technique Approaches

The difficulty level of a subject is needed either to understand the student acceptance of the subject and the highest level of student achievement in it. Some factors are considered, what kind of instructions, the readiness of the instructor and students in teaching and learning, evaluation and monitoring systems, and student expectations. Many factors are involved, and educators should know this. It is better if they can discern which are the prime factors and which the secondary factors. The purpose of the study is to find out the determinant factors in establishing the difficulty level of the subject from the students’, teachers’ and infrastructure point of view using three machine learning techniques. The MSE and the variable importance measurement were used to predict between some factors such as Attendance, Instructors, and other factors as independent variables and the difficulty level of the subject as a dependent variable. The study result showed that Gradient Boosting Machine obtained the MSE value result 1.14 and 1.30 for training and validation dataset. The model generated five variable importance as an independent factor, i.e. Attendance, Instructor, The course can give a new perspective to students, The quizzes, assignments, projects and exams contributed to helping the learning, and The Instructor was committed to the course and was understandable. The Gradient Boosting Machine is superior to other methods with the lowest MSE and MAE values results. Two methods, Gradient Boosting Machine and Deep Learning, have produced the same five main factors that influenced the difficulty of the subject. It means these factors are significant and should get intention by the stakeholders

Efficacy of Integrating a Novel 16-Gene Biomarker Panel and Intelligence Classifiers for Differential Diagnosis of Rheumatoid Arthritis and Osteoarthritis.

Introducing novel biomarkers for accurately detecting and differentiating rheumatoid arthritis (RA) and osteoarthritis (OA) using clinical samples is essential. In the current study, we searched for a novel data-driven gene signature of synovial tissues to differentiate RA from OA patients. Fifty-three RA, 41 OA, and 25 normal microarray-based transcriptome samples were utilized. The area under the curve random forests (RF) variable importance measurement was applied to seek the most influential differential genes between RA and OA. Five algorithms including RF, k-nearest neighbors (kNN), support vector machines (SVM), naïve-Bayes, and a tree-based method were employed for the classification. We found a 16-gene signature that could effectively differentiate RA from OA, including TMOD1, POP7, SGCA, KLRD1, ALOX5, RAB22A, ANK3, PTPN3, GZMK, CLU, GZMB, FBXL7, TNFRSF4, IL32, MXRA7, and CD8A. The externally validated accuracy of the RF model was 0.96 (sensitivity = 1.00, specificity = 0.90). Likewise, the accuracy of kNN, SVM, naïve-Bayes, and decision tree was 0.96, 0.96, 0.96, and 0.91, respectively. Functional meta-analysis exhibited the differential pathological processes of RA and OA; suggested promising targets for further mechanistic and therapeutic studies. In conclusion, the proposed genetic signature combined with sophisticated classification methods may improve the diagnosis and management of RA patients.

Predictive single-step kinetic model of biomass devolatilization for CFD applications: A comparison study of empirical correlations (EC), artificial neural networks (ANN) and random forest (RF)

The single-step model has been widely used for devolatilization in the computational fluid dynamics (CFD) of biomass gasification and combustion due to its low computational cost. The kinetic parameters of the single-step model are just obtained from previous studies and kept constant without regarding to the effects of biomass types and heating conditions, resulting in an obvious deviation on the treatment of biomass devolatilization. Here, several models, including the empirical correlations (EC), artificial neural networks (ANN) and random forest (RF) models, are developed to predict the kinetic parameters of the single-step model for CFD applications based on the biomass chemical compositions and heating conditions, and their performances are compared to highlight the optimal model. Two biomass devolatilization databases, used for training and validation respectively, are constructed from available experiments in the literature. The kinetic parameters are then fit from the database. The training and validation results show the EC model provides a poor performance with the lowest determination coefficients (R2≤0.80), while the ANN and RF models show an obviously better performance, with the ANN giving the middle performance (R2≥0.84) and the RF model giving the best performance (R2≥0.92). The variable importance measurement (VIM) results are also presented and discussed.

Estimation of coking indexes based on parental coal properties by variable importance measurement and boosted-support vector regression method

Coke as a fuel has a critical role for steel making industries. Since coke is a product of blended coals, it is essential to study relationships between parental coal components with quality of their coke products. Free swelling index (FSI) and maximum fluidity (MF) are standard coking indexes that widely used for blending coals and measuring quality of products. This study has been explored interdependencies between measured coal components by mutual information (MI) method and evaluated their importance in the prediction of coking indexes for a wide range of Illinois coal samples. MI results indicated that the set of moisture-organic sulfur and moisture-nitrogen-sulfate sulfur were the best variables for predictions of log(MF) and FSI, respectively. Adaptive Boosting method based on support vector regression (SVR), called Boosted-SVR, was used the selected variable sets for predictions of coking indexes. In testing stage of models, correlation of determination (R2) between actual and predicted values for the log(MF) and FSI were 0.89 and 0.90, respectively. These results indicated that Boosted-SVR model could quite satisfactory predict coking indexes. In general, outcomes of this investigation demonstrated an appropriate potential of coking quality prediction with limited numbers of input variables and suggested that a combination of MI with Boosted-SVR model as a new powerful tool which can be used for the computation of other complex fuel and processing problems based on measurement of conventional properties.

Rule-Based Intelligent System for Variable Importance Measurement and Prediction of Ash Fusion Indexes

Ash fusion temperatures [AFTs: initial deformation temperature (IDT), softening temperature (ST), and fluid temperature (FT)] are standard keys to estimate behavior of ash oxide for using coal and controlling the slag making at boilers. In this study, the modeling of AFTs based on ash oxide contents for 6537 U.S. coal samples have been investigated by a rule-based intelligent system (RBIS). Variable importance measurements (VIMs) of RBIS through the database indicated that Al2O3 contents in coal samples have the highest importance for prediction of AFTs. The RBIS model based on various rules was generated for predictions of IDT, ST, and FT. A comparison between RBIS and other typical predictive models [linear regression, genetic algorithm–neural network (GA–NN), and multilayer perceptron trained by back-propagation algorithm (MLP-BP)] was implemented to assess the capability of this purposed predictive model. Results indicated that RBIS can quite satisfactory predict AFTs, where R2 for IDT, ST, and FT for...

Study Relationship Between the Coal Thermoplastic Factor With Its Organic and Inorganic Properties by the Support Vector Regression Method

ABSTRACT Metallurgical cokes, as fuel for blast furnaces, have certain properties which are directly related to their blended parental coal characters. The maximum fluidity (MF) of coal as an energy index is typically used to estimate the coke properties. In this investigation, Support Vector Regression (SVR), as an intelligent method, has been applied to link characteristics and pyrolysis properties of coal samples with their representative MFs. SVR variable importance measurement (VIM) through a wide range of coal properties indicated that volatile matter (VM) and maximum vitrinite reflectance (Rmax) are the most effective parameters for the MF prediction. The results indicated that low rank coal samples (VM>45% and Rmax>0.7) have log(MF) higher than 14 and high rank ones (VM<35% and Rmax<0.6) have log(MF) less than 4. The evaluation of the SVR model trained with these two selected input variables showed that SVR can predict MF quite accurately where the coefficient of determination (R 2) between actual MF and SVR predicted was 0.86. According to these results, generation of SVR models which can predict and measure variable importance dependently, potentially may be applied for the scaling up of laboratory coal thermoplastic behavior to industrial levels, helping to sustainable development, and satisfactorily estimating coal consumption in the steel-making plants.