• The Journal of the Acoustical Society of Korea
• /
• v.26 no.5
• /
• pp.194-198
• /
• 2007

In this paper, we apply a discriminative weight training to a statistical model-based voice activity detection(VAD). In our approach, the VAD decision rule is expressed as the geometric mean of optimally weighted likelihood ratios(LRs) based on a minimum classification error(MCE) method which is different from the previous works in that different weights are assigned to each frequency bin which is considered more realistic. According to the experimental results, the proposed approach is found to be effective for the statistical model-based VAD using the LR test.

• Communications for Statistical Applications and Methods
• /
• v.26 no.6
• /
• pp.591-610
• /
• 2019

Deep Learning is one of the machine learning methods to find features from a huge data using non-linear transformation. It is now commonly used for supervised learning in many fields. In particular, Convolutional Neural Network (CNN) is the best technique for the image classification since 2012. For users who consider deep learning models for real-world applications, Keras is a popular API for neural networks written in Python and also can be used in R. We try examine the parameter estimation procedures of Deep Neural Network and structures of CNN models from basics to advanced techniques. We also try to figure out some crucial steps in CNN that can improve image classification performance in the CIFAR10 dataset using Keras. We found that several stacks of convolutional layers and batch normalization could improve prediction performance. We also compared image classification performances with other machine learning methods, including K-Nearest Neighbors (K-NN), Random Forest, and XGBoost, in both MNIST and CIFAR10 dataset.

• Communications for Statistical Applications and Methods
• /
• v.26 no.6
• /
• pp.635-644
• /
• 2019

Class probability is a fundamental target in classification that contains complete classification information. In this article, we propose a class probability estimation method when the predictor is functional. Motivated by Wang et al. (Biometrika, 95, 149-167, 2007), our estimator is obtained by training a sequence of functional weighted support vector machines (FWSVM) with different weights, which can be justified by the Fisher consistency of the hinge loss. The proposed method can be extended to multiclass classification via pairwise coupling proposed by Wu et al. (Journal of Machine Learning Research, 5, 975-1005, 2004). The use of FWSVM makes our method model-free as well as computationally efficient due to the piecewise linearity of the FWSVM solutions as functions of the weight. Numerical investigation to both synthetic and real data show the advantageous performance of the proposed method.

• Proceedings of the Korean Operations and Management Science Society Conference
• /
• 1996.10a
• /
• pp.31-34
• /
• 1996

The extensive researches have compared the performance of neural networks(NN) with those of various statistical techniques for the classification problem. The empirical results of these comparative studies have indicated that the neural networks often outperform the traditional statistical techniques. Moreover, there are some efforts that try to combine various classification methods, especially multivariate discriminant analysis with neural networks. While these efforts improve the performance, there exists a problem violating robust assumptions of multivariate discriminant analysis that are multivariate normality of the independent variables and equality of variance-covariance matrices in each of the groups. On the contrary, cluster analysis alleviates this assumption like neural networks. We propose a new approach to classification problems by combining the cluster analysis with neural networks. The resulting predictions of the composite model are more accurate than each individual technique.

• Journal of the Korean Statistical Society
• /
• v.26 no.4
• /
• pp.477-491
• /
• 1997

A new predictive classification rule for assigning future cases into one of two multivariate normal population (with unknown normal mixture model) is considered. The development involves calculation of posterior probability of each possible normal-mixture model via a default Bayesian test criterion, called intrinsic Bayes factor, and suggests predictive distribution for future cases to be classified that accounts for model uncertainty by weighting the effect of each model by its posterior probabiliy. In this paper, our interest is focused on constructing the classification rule that takes care of uncertainty about the types of covariance matrices (homogeneity/heterogeneity) involved in the model. For the constructed rule, a Monte Carlo simulation study demonstrates routine application and notes benefits over traditional predictive calssification rule by Geisser (1982).

• Communications for Statistical Applications and Methods
• /
• v.26 no.2
• /
• pp.79-89
• /
• 2019

Classification models pertaining to receiver operating characteristic (ROC) curve analysis have been extended from univariate to multivariate setup by linearly combining available multiple markers. One such classification model is the multivariate ROC curve analysis. However, not all markers contribute in a real scenario and may mask the contribution of other markers in classifying the individuals/objects. This paper addresses this issue by developing an algorithm that helps in identifying the important markers that are significant and true contributors. The proposed variable selection framework is supported by real datasets and a simulation study, it is shown to provide insight about the individual marker's significance in providing a classifier rule/linear combination with good extent of classification.

• Communications for Statistical Applications and Methods
• /
• v.22 no.6
• /
• pp.575-587
• /
• 2015

The development in data collection techniques results in high dimensional data sets, where discrimination is an important and commonly encountered problem that are crucial to resolve when high dimensional data is heterogeneous (non-common variance covariance structure for classes). An example of this is to classify microbial habitat preferences based on codon/bi-codon usage. Habitat preference is important to study for evolutionary genetic relationships and may help industry produce specific enzymes. Most classification procedures assume homogeneity (common variance covariance structure for all classes), which is not guaranteed in most high dimensional data sets. We have introduced regularized elimination in partial least square coupled with QDA (rePLS-QDA) for the parsimonious variable selection and classification of high dimensional heterogeneous data sets based on recently introduced regularized elimination for variable selection in partial least square (rePLS) and heterogeneous classification procedure quadratic discriminant analysis (QDA). A comparison of proposed and existing methods is conducted over the simulated data set; in addition, the proposed procedure is implemented to classify microbial habitat preferences by their codon/bi-codon usage. Five bacterial habitats (Aquatic, Host Associated, Multiple, Specialized and Terrestrial) are modeled. The classification accuracy of each habitat is satisfactory and ranges from 89.1% to 100% on test data. Interesting codon/bi-codons usage, their mutual interactions influential for respective habitat preference are identified. The proposed method also produced results that concurred with known biological characteristics that will help researchers better understand divergence of species.

• Proceedings of the KSRS Conference
• /
• 2002.10a
• /
• pp.388-393
• /
• 2002

Remote sensing data is collected and analyzed to enhance understanding of the terrestrial surface. Since Landsat satellite was launched in 1972, many researches using multispectral data has been achieved. Recently, with the availability of airborne and satellite hyperspectral data, the study on hyperspectral data are being increased. It is known that as the number of spectral bands of high-spectral resolution data increases, the ability to detect more detailed cases should also increase, and the classification accuracy should increase as well. In this paper, we classified the hyperspectral and multispectral data and tested the classification accuracy. The MASTER(MODIS/ASTER Airborne Simulator, 50channels, 0.4~13$\mu$m) and Landsat TM(7channels) imagery including Yeong-Gwang area were used and we adjusted the classification items in several cases and tested their classification accuracy through statistical comparison. As a result of this study, it is shown that hyperspectral data offer more information than multispectral data.

• The Korean Journal of Applied Statistics
• /
• v.19 no.3
• /
• pp.587-598
• /
• 2006

The number of variables exceeds the number of samples in microarray data. We propose a nonparametric local linear logistic classification procedure using orthogonal components for classifying high-dimensional microarray data. The proposed method is based on the local likelihood and can be applied to multi-class classification. We applied the local linear logistic classification method using PCA, PLS, and factor analysis components as new features to Leukemia data and colon data, and compare the performance of the proposed method with the conventional statistical classification procedures. The proposed method outperforms the conventional ones for each component, and PLS has shown best performance when it is embedded in the proposed method among the three orthogonal components.

• Communications for Statistical Applications and Methods
• /
• v.17 no.4
• /
• pp.527-540
• /
• 2010

Rule-based classification analysis is widely used for massive datamining because it is easy to understand and its algorithm is uncomplicated. In this classification analysis, majority vote of rules or weighted combination of rules using their supports are frequently used in order to combine rules. We propose a method to combine rules by using the multinomial distribution in this paper. Iterative proportional fitting algorithm is used to estimate the multinomial distribution which maximizes entropy constrained on rules' support. Simulation experiments show that this method can compete with other well known classification models in the case of two similar populations.