• Journal of Korea Society of Industrial Information Systems
• /
• v.17 no.7
• /
• pp.139-148
• /
• 2012

An ensemble of classifiers is to employ a set of individually trained classifiers and combine their predictions. It has been found that in most cases the ensembles produce more accurate predictions than the base classifiers. Combining outputs from multiple classifiers, known as ensemble learning, is one of the standard and most important techniques for improving classification accuracy in machine learning. An ensemble of classifiers is efficient only if the individual classifiers make decisions as diverse as possible. Bagging is the most popular method of ensemble learning to generate a diverse set of classifiers. Diversity in bagging is obtained by using different training sets. The different training data subsets are randomly drawn with replacement from the entire training dataset. The random subspace method is an ensemble construction technique using different attribute subsets. In the random subspace, the training dataset is also modified as in bagging. However, this modification is performed in the feature space. Bagging and random subspace are quite well known and popular ensemble algorithms. However, few studies have dealt with the integration of bagging and random subspace using SVM Classifiers, though there is a great potential for useful applications in this area. The focus of this paper is to propose methods for improving SVM performance using hybrid ensemble strategy for bankruptcy prediction. This paper applies the proposed ensemble model to the bankruptcy prediction problem using a real data set from Korean companies.

• Korean Journal of Air-Conditioning and Refrigeration Engineering
• /
• v.27 no.6
• /
• pp.283-292
• /
• 2015

A model of apartment heating load in a district heating system could be useful in the management and utilization of energy resources, since it could predict energy usage and so could assist in the efficient use of energy resources. The heating load in a district heating system varies in a highly nonlinear manner and is subject to many different factors, such as heating area, number of people living in that complex, and ambient temperature. Thus there are few published papers with accurate models of heating load, especially in domestic literature. This work is concerned with the modeling of apartment heating load in a district heating system in winter, using the reduced least square support vector machine (LS-SVM), and with the purpose of using the model to predict heating energy usage in domestic city area. We collected 23,856 pieces of data on heating energy usage over a 12-week period in winter, from 12 heat exchangers in five apartments. Half of the collected data were used to construct the heating load model, and the other half were used to test the model's accuracy. The model was able to predict the heating energy usage pattern rather accurately. It could also estimate the usage of heating energy within of mean absolute percentage error. This implies that the model prediction accuracy needs to be improved further, but it still could be considered as an acceptable model if we consider the nonlinearity and uncertainty of apartment heating energy usage in a district heating system.

• Journal of Korea Multimedia Society
• /
• v.11 no.9
• /
• pp.1302-1309
• /
• 2008

Traditional clustering methods, like k-means or fuzzy clustering, are prototype-based methods which are applicable only to convex clusters. On the other hand, spectral clustering tries to find clusters only using local similarity information. Its ability to handle concave clusters has gained the popularity recent years together with support vector machine (SVM) which is a kernel-based classification method. However, as is in SVM, the kernel width plays an important role and has a great impact on the result. Several methods are proposed to decide it automatically, it is still determined based on heuristics. In this paper, we proposed an adaptive method deciding the kernel width based on distance histogram. The proposed method is motivated by the fact that the affinity matrix should be formed into a block diagonal matrix to generate the best result. We use the tradition Euclidean distance together with the random walk distance, which make it possible to form a more apparent block diagonal affinity matrix. Experimental results show that the proposed method generates more clear block structured affinity matrix than the existing one does.

• KIPS Transactions on Software and Data Engineering
• /
• v.5 no.1
• /
• pp.43-50
• /
• 2016

This paper introduces an appearance-based facial expression recognition method using ASM landmarks which is used to acquire a detailed face region. In particular, EHMM-based algorithm and SVM classifier with histogram feature are employed to appearance-based facial expression recognition, and performance evaluation of proposed method was performed with CK and JAFFE facial expression database. In addition, performance comparison was achieved through comparison with distance-based face normalization method and a geometric feature-based facial expression approach which employed geometrical features of ASM landmarks and SVM algorithm. As a result, the proposed method using ASM-based face normalization showed performance improvements of 6.39% and 7.98% compared to previous distance-based face normalization method for CK database and JAFFE database, respectively. Also, the proposed method showed higher performance compared to geometric feature-based facial expression approach, and we confirmed an effectiveness of proposed method.

• Journal of Digital Contents Society
• /
• v.19 no.4
• /
• pp.821-827
• /
• 2018

Affective computing has gained increasing interest in the recent years with the development of potential applications in Human computer interaction (HCI) and healthcare. Although momentous research has been done on human emotion recognition, however, in comparison to speech and facial expression less attention has been paid to physiological signals. In this paper, Electroencephalogram (EEG) signals from different brain regions were investigated using modified wavelet energy features. For minimization of redundancy and maximization of relevancy among features, mRMR algorithm was deployed significantly. EEG recordings of a publically available "DEAP" database have been used to classify four classes of emotions with Multi class Support Vector Machine. The proposed approach shows significant performance compared to existing algorithms.

• Journal of Korea Game Society
• /
• v.16 no.2
• /
• pp.7-16
• /
• 2016

In this paper, we propose a method for effectively detecting specific behavior. The proposed method detects outlying behavior based on the game players' characteristics. These characteristics are captured non-invasively in a general game environment and add keystroke based on repeated pattern. In this paper, cameras were used to analyze observed data such as facial expressions and player movements. Moreover, multimodal data from the game players was used to analyze high-dimensional game-player data for a detection effect of repeated behaviour pattern. A support vector machine was used to efficiently detect outlying behaviors. We verified the effectiveness of the proposed method using games from several genres. The recall rate of the outlying behavior pre-identified by industry experts was approximately 70%. In addition, Repeated behaviour pattern can be analysed possible. The proposed method can also be used for feedback and quantification about analysis of various interactive content provided in PC environments.

• Atmosphere
• /
• v.20 no.4
• /
• pp.415-426
• /
• 2010

A Land cover map over East Asian region (Kongju national university Land Cover map: KLC) is classified by using support vector machine (SVM) and evaluated with ground truth data. The basic input data are the recent three years (2006-2008) of MODIS (MODerate Imaging Spectriradiometer) NDVI (normalized difference vegetation index) data. The spatial resolution and temporal frequency of MODIS NDVI are 1km and 16 days, respectively. To minimize the number of cloud contaminated pixels in the MODIS NDVI data, the maximum value composite is applied to the 16 days data. And correction of cloud contaminated pixels based on the spatiotemporal continuity assumption are applied to the monthly NDVI data. To reduce the dataset and improve the classification quality, 9 phenological data, such as, NDVI maximum, amplitude, average, and others, derived from the corrected monthly NDVI data. The 3 types of land cover maps (International Geosphere Biosphere Programme: IGBP, University of Maryland: UMd, and MODIS) were used to build up a "quasi" ground truth data set, which were composed of pixels where the three land cover maps classified as the same land cover type. The classification results show that the fractions of broadleaf trees and grasslands are greater, but those of the croplands and needleleaf trees are smaller compared to those of the IGBP or UMd. The validation results using in-situ observation database show that the percentages of pixels in agreement with the observations are 80%, 77%, 63%, 57% in MODIS, KLC, IGBP, UMd land cover data, respectively. The significant differences in land cover types among the MODIS, IGBP, UMd and KLC are mainly occurred at the southern China and Manchuria, where most of pixels are contaminated by cloud and snow during summer and winter, respectively. It shows that the quality of raw data is one of the most important factors in land cover classification.

• Journal of the Korean Data and Information Science Society
• /
• v.24 no.6
• /
• pp.1341-1348
• /
• 2013

To solve the classification problems, various data mining techniques have been applied to database marketing, credit scoring and market forecasting. In this paper, we compare various techniques such as bagging, boosting, LASSO, random forest and support vector machine with the daily lens transaction data. The classical techniques-decision tree, logistic regression-are used too. The experiment shows that the random forest has a little smaller misclassification rate and standard error than those of other methods. The performance of the SVM is good in the sense of misclassfication rate and bad in the sense of standard error. Taking the model interpretation and computing time into consideration, we conclude that the LASSO gives the best result.

• Journal of Korea Multimedia Society
• /
• v.6 no.5
• /
• pp.788-796
• /
• 2003

This paper proposes using dynamic momentum for squential learning method. Using The dynamic momentum improves convergence speed and performance by the variable momentum, also can identify it in the RMSE(root mean squared error). The proposed method is reflected using variable momentum according to current state. While static momentum is equally influenced on the whole, dynamic momentum algorithm can control the convergence rate and performance. According to the variable change of momentum by training. Unlike former classification and regression problems, this paper confirms both performance and regression rate of the dynamic momentum. Using RMSE(root mean square error ), which is one of the regression methods. The proposed dynamic momentum has been applied to the kernel adatron and kernel relaxation as the new sequential learning method of support vector machine presented recently. In order to show the efficiency of the proposed algorithm, SONAR data, the neural network classifier standard evaluation data, are used. The simulation result using the dynamic momentum has a better convergence rate, performance and RMSE than those using the static moment, respectively.

• Genomics & Informatics
• /
• v.5 no.3
• /
• pp.102-106
• /
• 2007

Radiofrequency (RF) radiation at the frequency of mobile phones has been not reported to induce cellular responses in in vitro and in vivo models. We exposed HEI-OC1, conditionally-immortalized mouse auditory cells, to RF radiation to characterize cellular responses to 1763 MHz RF radiation. While we could not detect any differences upon RF exposure, whole-genome expression profiling might provide the most sensitive method to find the molecular responses to RF radiation. HEI-OC1 cells were exposed to 1763 MHz RF radiation at an average specific absorption rate (SAR) of 20 W/kg for 24 hr and harvested after 5 hr of recovery (R5), alongside sham-exposed samples (S5). From the whole-genome profiles of mouse neurons, we selected 9 differentially-expressed genes between the S5 and R5 groups using information gain-based recursive feature elimination procedure. Based on support vector machine (SVM), we designed a prediction model using the 9 genes to discriminate the two groups. Our prediction model could predict the target class without any error. From these results, we developed a prediction model using biomarkers to determine the RF radiation exposure in mouse auditory cells with perfect accuracy, which may need validation in in vivo RF-exposure models.