• The Journal of the Korea Contents Association
• /
• v.21 no.11
• /
• pp.135-144
• /
• 2021

Recently, data-driven decision-making technology has become a key technology leading the data industry, and machine learning technology for this requires high-quality training datasets. However, real-world data contains missing values for various reasons, which degrades the performance of prediction models learned from the poor training data. Therefore, in order to build a high-performance model from real-world datasets, many studies on automatically imputing missing values in initial training data have been actively conducted. Many of conventional machine learning-based imputation techniques for handling missing data involve very time-consuming and cumbersome work because they are applied only to numeric type of columns or create individual predictive models for each columns. Therefore, this paper proposes a new data imputation technique called 'Denoising Self-Attention Network (DSAN)', which can be applied to mixed-type dataset containing both numerical and categorical columns. DSAN can learn robust feature expression vectors by combining self-attention and denoising techniques, and can automatically interpolate multiple missing variables in parallel through multi-task learning. To verify the validity of the proposed technique, data imputation experiments has been performed after arbitrarily generating missing values for several mixed-type training data. Then we show the validity of the proposed technique by comparing the performance of the binary classification models trained on imputed data together with the errors between the original and imputed values.

• KIPS Transactions on Software and Data Engineering
• /
• v.11 no.3
• /
• pp.107-114
• /
• 2022

In this paper, we have enhanced the risk prediction of hypertension using the feature selection method in the Korean National Health and Nutrition Examination Survey (KNHANES) database of the Korea Centers for Disease Control and Prevention. The study identified various risk factors correlated with chronic hypertension. The paper is divided into three parts. Initially, the data preprocessing step of removes missing values, and performed z-transformation. The following is the feature selection (FS) step that used a factor analysis (FA) based on the feature selection method in the dataset, and feature importance (FI) and multicollinearity analysis (MC) were compared based on FS. Finally, in the predictive analysis stage, it was applied to detect and predict the risk of hypertension. In this study, we compare the accuracy, f-score, area under the ROC curve (AUC), and mean standard error (MSE) for each model of classification. As a result of the test, the proposed MC-FA-RF model achieved the highest accuracy of 80.12%, MSE of 0.106, f-score of 83.49%, and AUC of 85.96%, respectively. These results demonstrate that the proposed MC-FA-RF method for hypertension risk predictions is outperformed other methods.

• Journal of Information Processing Systems
• /
• v.16 no.1
• /
• pp.6-29
• /
• 2020

Biometrics identification using multiple modalities has attracted the attention of many researchers as it produces more robust and trustworthy results than single modality biometrics. In this paper, we present a novel multimodal recognition system that trains a deep learning network to automatically learn features after extracting multiple biometric modalities from a single data source, i.e., facial video clips. Utilizing different modalities, i.e., left ear, left profile face, frontal face, right profile face, and right ear, present in the facial video clips, we train supervised denoising auto-encoders to automatically extract robust and non-redundant features. The automatically learned features are then used to train modality specific sparse classifiers to perform the multimodal recognition. Moreover, the proposed technique has proven robust when some of the above modalities were missing during the testing. The proposed system has three main components that are responsible for detection, which consists of modality specific detectors to automatically detect images of different modalities present in facial video clips; feature selection, which uses supervised denoising sparse auto-encoders network to capture discriminative representations that are robust to the illumination and pose variations; and classification, which consists of a set of modality specific sparse representation classifiers for unimodal recognition, followed by score level fusion of the recognition results of the available modalities. Experiments conducted on the constrained facial video dataset (WVU) and the unconstrained facial video dataset (HONDA/UCSD), resulted in a 99.17% and 97.14% Rank-1 recognition rates, respectively. The multimodal recognition accuracy demonstrates the superiority and robustness of the proposed approach irrespective of the illumination, non-planar movement, and pose variations present in the video clips even in the situation of missing modalities.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.9 no.2
• /
• pp.600-619
• /
• 2015

We propose a cost-efficient hybrid algorithm for online location updates that efficiently combines feature point detection with the online trajectory-based sampling algorithm. Our algorithm is designed to minimize the average trajectory error with the minimal number of sample points. The algorithm is composed of 3 steps. First, we choose corner points from the map as sample points because they will most likely cause fewer trajectory errors. By employing the online trajectory sampling algorithm as the second step, our algorithm detects several missing and important sample points to prevent unwanted trajectory errors. The final step improves cost efficiency by eliminating redundant sample points on straight paths. We evaluate the proposed algorithm with real GPS trajectory data for various bus routes and compare our algorithm with the existing one. Simulation results show that our algorithm decreases the average trajectory error 28% compared to the existing one. In terms of cost efficiency, simulation results show that our algorithm is 29% more cost efficient than the existing one with real GPS trajectory data.

• Proceedings of the Korean Institute of Navigation and Port Research Conference
• /
• 2014.10a
• /
• pp.245-246
• /
• 2014

This paper suggests the navigation information display system which is based on augmented reality technology and especially focuses on image analysis technology. Navigator has to always confirm the information from marine electronic navigation devices and then they compare with the view of outside targets of the windows. During this 'head down' posture, they feel uncomfortable and sometimes it cause near-accidents such as collision or missing objects, because he or she cannot keep an eye on the front view of windows. Augmented reality can display both of information of virtual and real in a single display. Therefore we tried to adapt the AR technology to help navigators and have been studied and developed image pre-processing module as a previous research already. To analysis the outside view of the bridge window, we have extracted navigational information from the camera image by using image processing. This paper mainly describes about recognizing ship feature by haar-like feature and filtering region of interest area by AIS data, which are to improve accuracy of the image analysis.

• JSTS:Journal of Semiconductor Technology and Science
• /
• v.17 no.2
• /
• pp.288-293
• /
• 2017

Transient simulation capability with an implicit time derivation method is a missing feature in deterministic Boltzmann equation solvers. The H-transformation, which is critical for the stable simulation of nanoscale devices, introduces difficulties for the transient simulation. In this work, the transient simulation of graphene sheets is reported. It is shown that simulation of homogeneous systems can be done without abandoning the H-transformation, as much as a specially designed discretization method is employed. The AC mobility and step response of the graphene sheet on the $SiO_2$ substrate are simulated.

• Journal of Korea Multimedia Society
• /
• v.10 no.12
• /
• pp.1577-1584
• /
• 2007

Fingerprint verification is being adopted widely to provide positive identification with a high degree of confidence in all practical areas. This popular usage requires reliable methods for matching of these patterns. To meet the latest expectations, the paper presents a pair wise distance histogram method for fingerprint matching. Here, we introduced a randomized algorithm which exploits pair wise distances between the pairs of minutiae, as a basic feature for match. The method undergoes two steps for completion i.e. first it performs the matching locally then global matching parameters are calculated in second step. The proposed method is robust to common problems that fingerprint matching faces, such as scaling, rotation, translational changes and missing points etc. The paper includes the test of algorithm on various randomly generated minutiae and real fingerprints as well. The results of the tests resemble qualities and utility of method in related field.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.35 no.1
• /
• pp.47-56
• /
• 2012

Service Parts Management is defined as a supply management associated with service parts from the part suppliers to the final customer. A series of process to improve the customer service level by forecasting the demand and to minimize cost by maintaining the inventory level is included. Uniqueness such as missing value correction, the data pattern analysis and planned order system is designed and implemented. Main feature of order management system is to calculate order amount and order time based on selection of optimal forecasting algorithm.

• Asian-Australasian Journal of Animal Sciences
• /
• v.14 no.12
• /
• pp.1785-1793
• /
• 2001

The interaction of growth hormone with it's receptor results in dimerization of receptor, a feature known in action of certain cytokines. The interaction results in generation of number of signalling molecules. The involvement of Janus kinases, mitogen activated kinases, signal transduction and activator of transcription proteins, insulin like substrate, phosphatidylinositol 3-kinase, phospholipase C, protein kinase C is almost established in growth hormone action. There are still many missing links in explaining diversified activities of growth hormone. Amino acid sequence data for growth hormones and growth hormone receptors from a number of species have proved useful in understanding species specific effects of growth hormone. Complete understanding of growth hormone action can have implications in designing drugs for obtaining desired effects of growth hormone.

• Proceedings of the KOSOMBE Conference
• /
• v.1997 no.11
• /
• pp.75-78
• /
• 1997

In this paper we present new sound spectral analysis methods or prosthetic heart valve sounds. Phonocardiograms(PCG) of prosthetic heart valve were analyzed in order to derive frequency domain feature suitable or the classification of the valve state. The fast orthogonal search method and MUSIC (MUltiple SIgnal Classification) method are described or finding the significant frequencies in PCG. The fast orthogonal search method is effective with short data records and cope with noisy, missing and unequally-spaced data. MUSIC method's key to the performance is the division of the information in the autocorrelation matrix or the data matrix into two vector subspaces, one a signal subspace and the other a noise subspace.