Earticle

Occupational fatal injury rate per 10,000 population of Korea is still higher among the OECD member countries. To prevent fatal injuries, the causes of accidents including human error should be analyzed and then appropriate countermeasures should be established. There was an severe converter furnace accident resulting in five people death by chocking in 2013. Although the accident type of the furnace accident was suffocation, many safety problems were included before reaching the death of suffocation. If the safety problems are reviewed throughly, the alternative measures based on the review would be very useful in preventing similar accidents. In this study, we investigated the converter furnace accident by using human error analysis and accident scenario analysis. As a result, it was found that the accident was caused by some human errors, inappropriate task sequence and lack of control in coordinating work by several subordinating companies. From the review of this case, the followings are suggested: First, systematic human error analysis should be included in the investigation of fatal injury accidents. Second, multi man-machine accident scenario analyis is useful in most of coordinating work. Third, the more provision of information on system state will lessen human errors. Fourth, the coordinating control in safety should be performed in the work conducting by several different companies.

본 연구에서는 환경음 인식 성능의 향상을 위하여 GMM의 훈련 방식에 MCE 도입을 제안하였다. 이는 환경음 데이터 모델링에 사용할 분류오류함수를 정의할 때 해당 클래스의 로그우도 뿐 아니라 다른 클래스의 로그우도도 같이 고려함으로써 변별력 있는 분류가 이뤄질 수 있게 한다. 모델의 파라미터는 전체 클래스를 고려한 손실함수를 정의하고, GPD(generalized probabilistic descent)알고리즘을 이용하여 추정하였다. 제안된 방법의 인식 성능 비교를 위해 모두 9가지 환경음을 전처리 과정과 MFCC(mel-frequency cepstral coefficients)를 이용하여 12차 특징을 추출하고, 이를 혼합 성분의 수에 따라 GMM 분류 실험을 행하였다. 실험 결과에 따르면 혼합 성분을 19개 사용한 경우에서 MCE 훈련 방식이 평균 87.06%의 인식률로 가장 좋은 성능을 보였다. 이 결과로 제안한 MCE 훈련 방식이 환경음 인식에서 GMM의 훈련 방식으로 효과적으로 사용될 수 있음을 확인하였다.

In this paper, we proposed the MCE as a GMM training method to improve the performance of environmental sounds recognition. We model the environmental sounds data with newly defined misclassification function using the log likelihood of the corresponding class and the log likelihood of the rest classes for discriminative training. The model parameters are estimated with the loss function using GPD(generalized probabilistic descent). For recognition performance comparison, we extracted the 12 degrees features using preprocessing and MFCC(mel-frequency cepstral coefficients) of the 9 kinds of environmental sounds and carry out GMM classification experiments. According to the experimental results, MCE training method showed the best performance by an average of 87.06% with 19 mixtures. This result confirmed us that MCE training method could be effectively used as a GMM training method in environmental sounds recognition.

본 논문은 EC_NH의 설계와 구축을 설명한다. EC_NH는 멀티미디어 협동 작업 환경에서 소프트웨어 오류를 감지, 공유, 복구하기에 적합한 시스템이다. 이 시스템에 의해서 오류를 공유할 수 있다. 멀티미디어 공동 작업 환경의 관점에서 오류 공유는 협동 작업에 참가하는 참가자에게 상호작용적으로 오류를 공유한다. 디지털 캐로절은 사용자들에게 미디어 동기화 메카니즘을 통하여 미디어 객체 공유를 가능하게 한다. 본 시스템은 공동 작업에 참여한 사용자들이 다른 참여자들에게 같은 뷰로써 공유된 미디어 또는 오류 객체들을 참조할 수 있도록 구축하였다.

This paper explains the design and implementation of the EC_NH. EC_NH is a system that is suitable for detecting, sharing and recovering software error based on multimedia CSCW(Computer Supportes Cooperated Work). With error sharing system, a group cooperating users can share error applications. From the perspective of multimedia collaborative environment, an error application becomes another interactive presentation error is shared with participants engaged in a cooperative work. Our Digital Carousel enables user to share media objects through media synchronization mechanism. We implemented the Digital Carousel so that the users participated in collaborative work may refer shared media or error objects as the same view to others.

We develop methods for propagating and analyzing EPCI(Extended Process Capability Index) by using the error type that classifies into accuracy and precision. EPCI developed in this study can be applied to the three combined processes that consist of production, measurement and calibration. Little calibration work discusses while a great deal has been studied about SPC(Statistical Process Contol) and MSA(Measurement System Analysis). EPCI can be decomposed into three indexes such as PPCI(Production Process Capability Index), PPPI(Production Process Performance Index), MPCI(Measurement PCD, and CPCI(Calibration PCI). These indexs based on the type of error classification can be used with various statistical techniques and principles such as SPC control charts, ANOVA(Analysis of Variance), MSA Gage R&R, Additivity-of-Variance, and RSSM(Root Sum of Square Method). As the method proposed is simple, any engineer in charge of SPC. MSA and calibration can use efficientily in industries. Numerical examples are presentsed. We recommed that the indexes can be used in conjunction with evaluation criteria.

최근 기계 번역 기술과 자동 노이즈 생성 방법론을 기반으로 한국어 맞춤법 교정 연구가 활발히 이루어지고 있다. 해당 방법론들은 노이즈를 생성하여 학습 셋과 데이터 셋으로 사용한다. 이는 학습에 사용된 노이즈 외의 노이즈가 테스트 셋에 포함될 가능성이 낮아 정확한 성능 측정이 어렵다는 한계점이 존재한다. 또한 실제적인 오류 유형 분류 기준이 없어 연구마다 사용하는 오류 유형이 다르므로 질적 분석에 어려움을 겪고 있다. 이를 해결하기 위해 본 논문은 딥러닝 기반 한국어 맞춤법 교정 연구를 위한 새로운 ’오류 유형 분류 체계‘를 제안하며 이를 바탕으로 기존 상용화 한국어 맞춤법 교정기(시스템 A, 시스템 B, 시스템 C)에 대한 오류 분석을 수행하였다. 분석결과, 세 가지 교정 시스템들이 띄어쓰기 오류 외에 본 논문에서 제시한 다른 오류 유형은 교정을 잘 수행하지 못했으며 어순 오류나 시제 오류의 경우 오류 인식을 거의 하지 못함을 알 수 있었다.

Recently, studies on Korean spelling correction have been actively conducted based on machine translation and automatic noise generation. These methods generate noise and use as train and data set. This has limitation in that it is difficult to accurately measure performance because it is unlikely that noise other than the noise used for learning is included in the test set In addition, there is no practical error type standard, so the type of error used in each study is different, making qualitative analysis difficult. This paper proposes new 'error type classification' for deep learning-based Korean spelling correction research, and error analysis perform on existing commercialized Korean spelling correctors (System A, B, C). As a result of analysis, it was found the three correction systems did not perform well in correcting other error types presented in this paper other than spacing, and hardly recognized errors in word order or tense.

Clustering is the problem of identifying the distribution of patterns and intrinsic correlations in large data sets by partitioning the data points into similarity classes. In this paper, we study on the problem of clustering categorical data, where data objects are made up of non-numerical attributes. We propose MECC (Minimum Error Classification Clustering), an alternative technique for categorical data clustering using VPRS taking into account minimum error classification. The technique is implemented in MATLAB. Experimental results on two benchmark UCI datasets show that MECC technique is better than the baseline categorical data clustering techniques with respect to selecting the clustering attribute.

In discriminant analysis, we consider a special pattern which contains a block of missing observations. We assume that the two populations are equally likely and the costs of misclassification are equal. In this situation, we consider the bootstrap confidence intervals of the error rate in the circular models when the covariance matrices are equal and not equal.

In this paper, it will be assumed that there are two distinct populations which are multivariate normal with equal covariance matrix. We also assume that the two populations are equally likely and the costs of misclassification are equal. The classification rule depends on the situation when the training samples include missing values or not. We consider the bootstrap confidence intervals for classification error rate when a block of observation is missing.

In this paper, it will be assumed that there are two distinct populations which are multivariate normal with equal covariance matrix. We also assume that the two populations are equally likely and the costs of misclassification are equal. The classification rule depends on the situation whether the training samples include missing values or not. We consider the conditional bootstrap confidence intervals for classification error rate when a block of observation is missing.

본 논문에서는 잡음 환경에서 강인한 음성 인식을 위하여 특징 보상 기법의 성능을 향상시킬 수 있는 방법을 제안한다. 기존의 음성 모델 기반의 특징 보상 기법에서 이용되는 오염 음성 모델 추정 방식은 입력 음성에 대한 변별력 있는 사후 확률 예측을 보장하지 못하며, 부정확하게 계산된 사후 확률은 복구된 음성에서 명료도 하락의 문제를 일으킨다. 제안하는 기법에서는 오염 음성 모델 추정 과정에 분별적 훈련 방식의 하나인 최소 분류 오류 (MCE) 훈련 기법을 도입한다. MCE 훈련 기법을 적용하기 위해 변별력 하락의 가능성을 가지는 '경쟁 요소' 를 결정하는 기법을 제안한다. 병렬결합된 혼합 모델 (PCMM) 기반의 특징 보상에 MCE 훈련 기법을 적용하는 과정을 제안하고 변별력 향상의 영향을 관찰한다. Aurora 2.0 데이터베이스와 실제 자동차 주행 환경에서 수집된 음성 데이터베이스에 대한 성능 평가를 실시한다. 실험 결과는 제안한 기법이 음성 인식 성능 향상에 도움이 되는 것을 입증한다.

In this paper, we propose a scheme to improve discriminative property in the feature compensation method for robust speech recognition under noisy environments. The estimation of noisy speech model used in existing feature compensation methods do not guarantee the computation of posterior probabilities which discriminate reliably among the Gaussian components. Estimation of Posterior probabilities is a crucial step in determining the discriminative factor of the Gaussian models, which in turn determines the intelligibility of the restored speech signals. The proposed scheme employs minimum classification error (MCE) training for estimating the parameters of the noisy speech model. For applying the MCE training, we propose to identify and determine the 'competing components' that are expected to affect the discriminative ability. The proposed method is applied to feature compensation based on parallel combined mixture model (PCMM). The performance is examined over Aurora 2.0 database and over the speech recorded inside a car during real driving conditions. The experimental results show improved recognition performance in both simulated environments and real-life conditions. The result verifies the effectiveness of the proposed scheme for increasing the performance of robust speech recognition systems.

This study examined whether large language models (LLMs) can classify L2 learner errors in a way that agrees with human annotation. Three flagship LLMs ? GPT-5.5, Claude Opus 4.7, and Gemini 3.1 Pro Preview ? were tested on 1,000 instances from the preprocessed NUCLE corpus, covering the ten most frequent error types. Given the error span and the correction, each LLM predicted the error type. All three LLMs showed substantial agreement with the NUCLE gold labels (Cohen's κ = .78?.80) and almost perfect agreement with one another (Fleiss' κ = .923). However, systematic divergence appeared in four error types: Vt, Wform, Vform, and Wci. In a human evaluation of 60 cases, two independent reviewers selected the LLM-predicted label rather than the NUCLE gold label in 70.0% of the cases. This pattern suggests that the disagreements reflect plausible alternative readings allowed by NUCLE's taxonomy, rather than only simple LLM errors. For Gemini, different prompting strategies had little effect on performance. The findings suggest that multi-LLM consensus can serve as a diagnostic tool for overlapping error types, helping identify which labels are reliable and which require human review.

Statistics are proposed to assist in choosing the number of trees in random forests for classification, one of which is very useful in monitoring of the out-of-bag (OOB) error rate. These statistics are used sequentially as the forest is constructed and are computed from OOB votes or test set votes. They provide a measure of the expected disagreement between the current forest and the limiting forest with infinitely many trees. Examples and simulation studies of the statistic for monitoring of the OOB error rate are provided to illustrate the performance of these statistics.

The international nuclear industry has undergone a lot of changes since the Fukushima, Chernobyl and TMI nuclear power plant accidents. However, there are still large and small component deficiencies at nuclear power plants in the world. There are many causes of electrical equipment defects. There are also factors that cause component failures due to human errors. This paper analyzed the root causes of failure and types of human error in 300 cases of electrical component failures. We analyzed the operating experience of electrical components by methods of root causes in K-HPES (Korean-version of Human Performance Enhancement System) and by methods of human error types in HuRAM+ (Human error-Related event root cause Analysis Method Plus). As a result of analysis, the most electrical component failures appeared as circuit breakers and emergency generators. The major causes of failure showed deterioration and contact failure of electrical components by human error of operations management. The causes of direct failure were due to aged components. Types of human error affecting the causes of electrical equipment failure are as follows. The human error type group I showed that errors of commission (EOC) were 97%, the human error type group II showed that slip/lapse errors were 74%, and the human error type group III showed that latent errors were 95%. This paper is meaningful in that we have approached the causes of electrical equipment failures from a comprehensive human error perspective and found a countermeasure against the root cause. This study will help human performance enhancement in nuclear power plants. However, this paper has done a lot of research on improving human performance in the maintenance field rather than in the design and construction stages. In the future, continuous research on types of human error and prevention measures in the design and construction sector will be required.

Objectives: Different complex strategies of fusing handcrafted descriptors and features from convolutional neural network(CNN) models have been studied, mainly for two-class Papanicolaou (Pap) smear image classification. This paper explores asimplified system using combined binary coding for a five-class version of this problem. Methods: This system extracted featuresfrom transfer learning of AlexNet, VGG19, and ResNet50 networks before reducing this problem into multiple binarysub-problems using error-correcting coding. The learners were trained using the support vector machine (SVM) method. The outputs of these classifiers were combined and compared to the true class codes for the final prediction. Results: Despitethe superior performance of VGG19-SVM, with mean ± standard deviation accuracy and sensitivity of 80.68% ± 2.00% and80.86% ± 0.45%, respectively, this model required a long training time. There were also false-negative cases using both theVGGNet-SVM and ResNet-SVM models. AlexNet-SVM was more efficient in terms of running speed and prediction consistency. Our findings also showed good diagnostic ability, with an area under the curve of approximately 0.95. Further investigationalso showed good agreement between our research outcomes and that of the state-of-the-art methods, with specificityranging from 93% to 100%. Conclusions: We believe that the AlexNet-SVM model can be conveniently applied for clinicaluse. Further research could include the implementation of an optimization algorithm for hyperparameter tuning, as well asan appropriate selection of experimental design to improve the efficiency of Pap smear image classification.

In pattern classification, the Bhattacharyya distance has been used as a class separability measure and provides useful information for feature selection and extraction. In this paper, we propose a method to predict the classification error for multimodal data based on the Bhattacharyya distance. In our approach, we first approximate the pdf of multimodal distribution with a Gaussian mixture model and find the bhattacharyya distance and classification error. Exprimental results showed that there is a strong relationship between the Bhattacharyya distance and the classification error for multimodal data.

기존의 웹기반 가상실험 시스템은 단방향적 학습내용 제시 및 학습자의 반응이라는 구조를 가지고 있으며 실제 과학실험에서 범하기 쉬운 오류들을 학습자가 스스로 진단하고 수정할 수 있는 과정을 간과하고 있다. 또한 현재의 가상실험 사이트는 학습자의 입장에서 다양하게 사고하고 변인을 조작하며 피드백을 통한 오류를 분석, 분류 및 수정 할 수 있는 기회를 제공하지 않는 경우가 대부분이다. 이에 본 논문은 가상 실험에서 학습자들이 겪게 되는 오류발생 문제를 진단하고 분석하여 피드백과정을 거쳐 오류를 수정하고 보완하여 가상공간에서 교사와 학습자, 학습자와 학습자간의 상호작용을 통한 가상실험이 원활하게 이루어지게 하는데 목적이 있다. 본 논문에서는 웹 기반 가상실험 학습의 장점 및 그와 연관된 오류분석 및 피드백의 의미를 7차 교육과정과 관련되어 기술하였으며, 시스템을 인터페이스 프로세스, 오류체크 프로세스, 오류분석 프로세스, 오류기록 프로세스, 오류제어 프로세스, 피드백 프로세스로 구성하였다. 또한 현 교육과정 중 초등학교 4학년 과학과의 한 단원을 선택하여 시스템을 적용하였다. 먼저 국내 및 국외에서 웹기반 학습과 관련된 가상실험 사이트들이 다수 개발되고 있으나 단순한 학습내용 제시와 실험결과 위주의 평가를 중시한다는 단점이 있다. 본 시스템의 학습자는 이와는 달리 실험결과 뿐만 아니라 실험과정의 오류를 확인하고 피드백을 받아 스스로 오류를 분석, 수정하게 된다.

결론 및 제언

Tax noncompliance, generally called unfaithful report, is an important matter in the field of customs duties as well as internal taxes. Customs tax noncompliance mainly happens in regards to customs classification or customs valuation. In the field of customs classification, noncompliance more often arises from negligence or ignorance than by intention. Especially, when a product might be classified in two or more headings of the tariff schedule by multiple criteria, taxpayers are likely to make mistakes about customs classification.The researcher tried to search a suitable judicial case, reorganize it as this study purpose, and analyze it by the method of explanation building. With the pre-developed research questions, the researcher sought to find the reasons 'why' the major problems occurred, and the ways 'how' to solve them. Therefore, the purposes of this paper were to analyze the case containing the customs classification error which occurred by negligence of the principle of tariff classification, and to give some suggestions for both customs authorities and taxpayers.

클러스터링은 대표적인 비교사 학습 방법의 하나로 균일한 특성을 가지는 데이터를 군집으로 묶기 위해 사용된다. 균일한 특성을 가지는 데이터 부분집합을 문맥으로 정의하고 문맥 내에서 국부적으로 분류를 행하는 융합 방법이 사용되고 있지만 클러스터링은 비교사 학습 방법이라는 한계로 인해 클러스터링 결과로 만들어지는 문맥이 분류에 있어 최선임을 보장하기 어렵다. 이 논문에서는 생성된 클러스터를 문맥으로 가정하고 각 문맥에서 분류를 시행하는 경우 최소의 오류를 보일 수 있는, 분류를 고려한 클러스터링 기법을 제안한다. 제안하는 방법은 선형 판별 분석에서와 유사하게 클러스터 내 동일한 클래스에 속하는 데이터 쌍은 작은 거리 값을, 서로 다른 클래스에 속하는 데이터 쌍은 큰 거리 값을 가지도록 하기 위한 제약 조건을 적용하여 분류 오류를 줄이도록 하였다. 제안한 방법의 실효성은 실험 결과를 통해 확인할 수 있다.

Clustering is one of the most popular unsupervised learning methods, which is widely used to form clusters with homogeneous data. Clustering was used to extract contexts corresponding to clusters and a classification method was applied to each context or cluster individually. However, it is difficult to say that the unsupervised clustering is the best context forming method from the view of classification. In this paper, a new clustering method considering classification was proposed. The proposed method tries to minimize classification error in each cluster when a classification method is applied to each context locally. For this purpose, the proposed method adds constraints forcing two data points belong to the same class to have small distances, and two data points belong to different classes to have large distances in each cluster like in linear discriminant analysis. The usefulness of the proposed method is confirmed by experimental results.

영상분류 결과는 지형적 영향, 영상의 상태 등에 따라 전체 영상에 대하여 균일하지 않을 수 있다. 본 연구에서는 분류 결과의 불균일성과 위성영상 분류 오차의 공간적 분포를 평가하기 위해 ISDd (Index of Spatial Distribution by distance) 와 ISDs (ISD by scatteredness)의 개념을 제안하였다. ISDd는 지표화된 오분류 화소간의 거리이고, ISDs는 오분류 화소의 산포도에 관한 통계적 지표이다. 실제 위성영상에 대한 실험을 통하여 ISDd와 ISDs를 계산 및 평가하였으며, 실제 국지적 오분류 영역을 추출하여 오분류의 원인을 고찰하였다. 본 연구 결과, ISDd와 ISDs를 동시에 사용하여 오분류 화소의 국지적 밀집 여부와 밀집 정도의 평가가 가능하였으며, 그 결과를 토대로 영상의 일부분에 대한 분류결과의 채택/기각을 결정할 수 있었다. 따라서, 전체 분류정확도 외에 공간분포지표를 사용함으로써 사용자는 오분류 화소의 공간적 분포 상태를 파악할 수 있으며, 분류 결과의 적합성 및 신뢰성 판단을 위한 추가적인 기준을 가질 수 있다.

The quality of image classification results is not always uniform over entire image. Thus, this study proposes the concept of ISDd (Index of Spatial Distribution by distance) and ISDs (ISD by scatteredness) for the evaluation of unevenness of result quality, and spatial distribution of satellite image classification errors. The ISDd is indexed mean distance of misclassified pixels and the ISDs is statistical indicator of scatteredness of misclassified pixels. In this study, the ISDd and the ISDs are calculated and evaluated for some satellite images, then misclassified area is extracted and the reasons of misclassification are examined. As the result of this study, using both the ISDd and the ISDs, the basis of decision on adoption/rejection of classification results is offered at sub-image level by evaluation of the local aggregation of misclassified pixels. Using Index of Spatial Distribution. as well as overall classification accuracy, users can understand the spatial distribution of misclassified pixels, and can have the additional criterion of the judgement on suitability and reliability of classification results.