• Korean Journal of Radiology
• /
• v.22 no.3
• /
• pp.334-343
• /
• 2021

Objective: We aimed to develop a prediction model for diagnosing severe aortic stenosis (AS) using computed tomography (CT) radiomics features of aortic valve calcium (AVC) and machine learning (ML) algorithms. Materials and Methods: We retrospectively enrolled 408 patients who underwent cardiac CT between March 2010 and August 2017 and had echocardiographic examinations (240 patients with severe AS on echocardiography [the severe AS group] and 168 patients without severe AS [the non-severe AS group]). Data were divided into a training set (312 patients) and a validation set (96 patients). Using non-contrast-enhanced cardiac CT scans, AVC was segmented, and 128 radiomics features for AVC were extracted. After feature selection was performed with three ML algorithms (least absolute shrinkage and selection operator [LASSO], random forests [RFs], and eXtreme Gradient Boosting [XGBoost]), model classifiers for diagnosing severe AS on echocardiography were developed in combination with three different model classifier methods (logistic regression, RF, and XGBoost). The performance (c-index) of each radiomics prediction model was compared with predictions based on AVC volume and score. Results: The radiomics scores derived from LASSO were significantly different between the severe AS and non-severe AS groups in the validation set (median, 1.563 vs. 0.197, respectively, p < 0.001). A radiomics prediction model based on feature selection by LASSO + model classifier by XGBoost showed the highest c-index of 0.921 (95% confidence interval [CI], 0.869-0.973) in the validation set. Compared to prediction models based on AVC volume and score (c-indexes of 0.894 [95% CI, 0.815-0.948] and 0.899 [95% CI, 0.820-0.951], respectively), eight and three of the nine radiomics prediction models showed higher discrimination abilities for severe AS. However, the differences were not statistically significant (p > 0.05 for all). Conclusion: Models based on the radiomics features of AVC and ML algorithms may perform well for diagnosing severe AS, but the added value compared to AVC volume and score should be investigated further.

• Journal of the Korean Data and Information Science Society
• /
• v.21 no.3
• /
• pp.561-567
• /
• 2010

Multiclassification is typically performed using voting scheme methods based on combining a set of binary classifications. In this paper we use multiclassification method with a hat matrix of least squares support vector machine (LS-SVM), which can be regarded as the revised one-against-all method. To tackle multiclass problems for large data, we use the $Nystr\ddot{o}m$ approximation and the quadratic Renyi entropy with estimation in the primal space such as used in xed size LS-SVM. For the selection of hyperparameters, generalized cross validation techniques are employed. Experimental results are then presented to indicate the performance of the proposed procedure.

• International Journal of Contents
• /
• v.17 no.4
• /
• pp.79-90
• /
• 2021

With the deepening of population aging, pension has become an urgent problem in most countries. Community smart pension can effectively resolve the problem of traditional pension, as well as meet the personalized and multi-level needs of the elderly. To predict the pension intention of the elderly in the community more accurately, this paper uses the decision tree classification method to classify the pension data. After missing value processing, normalization, discretization and data specification, the discretized sample data set is obtained. Then, by comparing the information gain and information gain rate of sample data features, the feature ranking is determined, and the C4.5 decision tree model is established. The model performs well in accuracy, precision, recall, AUC and other indicators under the condition of 10-fold cross-validation, and the precision was 89.5%, which can provide the certain basis for government decision-making.

• Journal of the Korean Data and Information Science Society
• /
• v.20 no.2
• /
• pp.261-272
• /
• 2009

Kolmogorov-Smirnov (K-S) statistic has been widely used for testing homogeneity of two distributions in the credit rating models. Joseph (2005) used K-S statistic to obtain validation criteria which is most well-known. There are other homogeneity test statistics such as the Cramer-von Mises, Anderson-Darling, and Watson statistics. In this paper, these statistics are introduced and applied to obtain criterion of these statistics by extending Joseph (2005)'s work. Another set of alternative criterion is suggested according to various sample sizes, type a error rates, and the ratios of bads and goods by using the simulated data under the similar situation as real credit rating data. We compare and explore among Joseph's criteria and two sets of the proposed criterion and discuss their applications.

• Near Infrared Analysis
• /
• v.1 no.1
• /
• pp.1-7
• /
• 2000

A convenient algorithm for optimizing wavelength selection in multiple linear regression (MLR) has been developed. MOP (MLP Optimization Program) has been developed to test all possible MLR calibration models in a given spectral range and finally find an optimal MLR model with external validation capability. MOP generates all calibration models from all possible combinations of wavelength, and simultaneously calculates SEC (Standard Error of Calibration) and SEV (Standard Error of Validation) by predicting samples in a validation data set. Finally, with determined SEC and SEV, it calculates another parameter called SAD (Sum of SEC, SEV, and Absolute Difference between SEC and SEV: sum(SEC+SEV+Abs(SEC-SEV)). SAD is an useful parameter to find an optimal calibration model without over-fitting by simultaneously evaluating SEC, SEV, and difference of error between calibration and validation. The calibration model corresponding to the smallest SAD value is chosen as an optimum because the errors in both calibration and validation are minimal as well as similar in scale. To evaluate the capability of MOP, the determination of benzene content in unleaded gasoline has been examined. MOP successfully found the optimal calibration model and showed the better calibration and independent prediction performance compared to conventional MLR calibration.

• Coupled systems mechanics
• /
• v.4 no.1
• /
• pp.37-65
• /
• 2015

Modeling and simulation of mechanical response of infrastructure object, solids and structures, relies on the use of computational models to foretell the state of a physical system under conditions for which such computational model has not been validated. Verification and Validation (V&V) procedures are the primary means of assessing accuracy, building confidence and credibility in modeling and computational simulations of behavior of those infrastructure objects. Validation is the process of determining a degree to which a model is an accurate representation of the real world from the perspective of the intended uses of the model. It is mainly a physics issue and provides evidence that the correct model is solved (Oberkampf et al. 2002). Our primary interest is in modeling and simulating behavior of porous particulate media that is fully saturated with pore fluid, including cyclic mobility and liquefaction. Fully saturated soils undergoing dynamic shaking fall in this category. Verification modeling and simulation of fully saturated porous soils is addressed in more detail by (Tasiopoulou et al. 2014), and in this paper we address validation. A set of centrifuge experiments is used for this purpose. Discussion is provided assessing the effects of scaling laws on centrifuge experiments and their influence on the validation. Available validation test are reviewed in view of first and second order phenomena and their importance to validation. For example, dynamics behavior of the system, following the dynamic time, and dissipation of the pore fluid pressures, following diffusion time, are not happening in the same time scale and those discrepancies are discussed. Laboratory tests, performed on soil that is used in centrifuge experiments, were used to calibrate material models that are then used in a validation process. Number of physical and numerical examples are used for validation and to illustrate presented discussion. In particular, it is shown that for the most part, numerical prediction of behavior, using laboratory test data to calibrate soil material model, prior to centrifuge experiments, can be validated using scaled tests. There are, of course, discrepancies, sources of which are analyzed and discussed.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.11
• /
• pp.13-18
• /
• 2022

In this paper, we proposed VW-FedAVG(Validation based Weighted FedAVG) which updates the global model by weighting according to performance verification from the models of each device participating in the training. The first method is designed to validate each local client model through validation dataset before updating the global model with a server side validation structure. The second is a client-side validation structure, which is designed in such a way that the validation data set is evenly distributed to each client and the global model is after validation. MNIST, CIFAR-10 is used, and the IID, Non-IID distribution for image classification obtained higher accuracy than previous studies.

• Bulletin of the Korean Chemical Society
• /
• v.32 no.7
• /
• pp.2397-2404
• /
• 2011

Variable selection k nearest neighbor QSAR modeling approach was applied to a data set of 80 3-arylisoquinolines exhibiting cytotoxicity against human lung tumor cell line (A-549). All compounds were characterized with molecular topology descriptors calculated with the MolconnZ program. Seven compounds were randomly selected from the original dataset and used as an external validation set. The remaining subset of 73 compounds was divided into multiple training (56 to 61 compounds) and test (17 to 12 compounds) sets using a chemical diversity sampling method developed in this group. Highly predictive models characterized by the leave-one out cross-validated $R^2$ ($q^2$) values greater than 0.8 for the training sets and $R^2$ values greater than 0.7 for the test sets have been obtained. The robustness of models was confirmed by the Y-randomization test: all models built using training sets with randomly shuffled activities were characterized by low $q^2{\leq}0.26$ and $R^2{\leq}0.22$ for training and test sets, respectively. Twelve best models (with the highest values of both $q^2$ and $R^2$) predicted the activities of the external validation set of seven compounds with $R^2$ ranging from 0.71 to 0.93.

• Journal of Information Processing Systems
• /
• v.14 no.5
• /
• pp.1075-1086
• /
• 2018

In this paper a cross-validation algorithm for training probabilistic neural networks (PNNs) is presented in order to be applied to automatic face identification. Actually, standard PNNs perform pretty well for small and medium sized databases but they suffer from serious problems when it comes to using them with large databases like those encountered in biometrics applications. To address this issue, we proposed in this work a new training algorithm for PNNs to reduce the hidden layer's size and avoid over-fitting at the same time. The proposed training algorithm generates networks with a smaller hidden layer which contains only representative examples in the training data set. Moreover, adding new classes or samples after training does not require retraining, which is one of the main characteristics of this solution. Results presented in this work show a great improvement both in the processing speed and generalization of the proposed classifier. This improvement is mainly caused by reducing significantly the size of the hidden layer.

• Korean Management Science Review
• /
• v.23 no.1
• /
• pp.63-74
• /
• 2006

Decision tree induction algorithm is one of the most widely used methods in classification problems. However, they could be trapped into a local minimum and have no reasonable means to escape from it if tree algorithm uses top-down search algorithm. Further, if irrelevant or redundant features are included in the data set, tree algorithms produces trees that are less accurate than those from the data set with only relevant features. We propose a hybrid algorithm to generate decision tree that uses genetic programming with sequentially selected features. Correlation-based Feature Selection (CFS) method is adopted to find relevant features which are fed to genetic programming sequentially to find optimal trees at each iteration. The new proposed algorithm produce simpler and more understandable decision trees as compared with other decision trees and it is also effective in producing similar or better trees with relatively smaller set of features in the view of cross-validation accuracy.