Earticle

확장된 데이터 표현의 주요 목표는 유비쿼터스 환경에서 일반적인 문제에 적합한 데이터 구조를 개발하는 것이다. 이 방법의 가장 큰 특징은 속성 값을 확률로 표현할 수 있다는 것이다. 다음 특성은 훈련 데이터의 각 이벤트가 중요도를 나타내는 가중치 값을 갖도록 한다는 것이다. 데이터 구조가 개발된 후에 이를 학습할 수 있는 알고리즘이 고안된다. 그 동안 이 알고리즘은 여러 분야에서 여러 문제에 적용하여 좋은 결과를 산출해 왔다. 본 논문은 먼저 데이터 표현 기법인 UChoo를 소개하고 이론적인 배경이 되는 규칙 개선 문제를 소개한다. 그리고 규칙 개선, 손실 데이터 처리, BEWS 문제, 앙상블 시스템과 같은 응용 분야의 예를 소개한다.

The main goal of extended data expression is to develop a data structure suitable for common problems in ubiquitous environments. The greatest feature of this method is that the attribute values can be represented with probability. The next feature is that each event in the training data has a weight value that represents its importance. After this data structure has been developed, an algorithm has been devised that can learn it. In the meantime, this algorithm has been applied to various problems in various fields to obtain good results. This paper first introduces the extended data expression technique, UChoo, and rule refinement method, which are the theoretical basis. Next, this paper introduces some examples of application areas such as rule refinement, missing data processing, BEWS problem, and ensemble system.

본 연구에서는 바둑학습에서의 중요한 부분인 사활문제의 분류체계를 정립하고자 하였다. 기존의 사활문제집에서는 사활문제에 대한 체계적인 분류체계가 정립되어 있지 않은 관계로 문제의 구성과 체계에 있어서 난삽함을 노정하고 있다. 이러한 문제점은 결국 사활 학습의 어려움을 초래 한다. 본 연구에서는 사활의 분류체계를 7가지 하위 차원으로 구분하고, 이러한 분류체계에 따라 사활문제집의 구성을 제안하였다.

The study tried to establish the classification system for life and death problems in Go. The classification system is composed of 7 sub-dimensions, such as level of difficulty, eye space, method of solving, location of board, dual life, race, ko. It is very important to make life and death problems adjusted to level of learner. Decomposing problems into sub-dimensions would make learning of life and death problems much easier. The study proposed the method of composing the problems according to several dimensions.

본 연구에서는 성인의 비자살적 자해 잠재계층유형을 확인하고, 잠재계층유형에 영향을 미치는 위험요인과 잠재계층유형에 따른 정서 및 행동문제의 차이를 파악하고자 하였다. 이를 위해 전국의 성인을 대상으로 온라인 설문조사를 통해 수집된 자료와 오프라인 설문조사를 통해 수집된 총 188명의 자료를 분석에 사용하였다. 연구결과, 성인의 비자살적 자해의 잠재계층은 ‘경미형’, ‘중도형’, ‘심각형’의 집단으로 확인되었다. 다음으로 성인의 비자살적 자해의 잠재계층유형의 분류에 영향을 미치는 위험요인을 살펴본 결과, 연령이 낮을수록 경미형보다 중도형 집단에, 중도형보다 심각형 집단에 속할 가능성이 높고, 아동기 학대를 경험하고, 고통감내력이 저하될수록 경미형 및 중도형보다 심각형 집단에 속할 가능성이 높았으며, 대인관계를 통해 자해에 대해 노출되었을수록 경미형보다 심각형 집단에 속할 가능성이 높은 것으로 나타났다. 마지막으로 성인의 비자살적 자해의 잠재계층유형에 따른 정서 및 행동문제의 차이를 살펴본 결과, 심각형 집단의 우울수준이 가장 높았고, 다음은 중도형, 경미형 집단 순으로 나타났다. 심각형 집단은 경미형 집단에 비해 불안수준이 더 높은 것으로 나타났고, 심각형 집단은 경미형과 중도형 집단 각각에 비해 폭식의 수준이 더 높은 것으로 나타났다. 연구결과를 토대로 논의 및 시사점을 제시하였다.

The purpose of this study was to classify adults’ self-injurious behavior using latent classification analysis, to identify risk factors affecting latent class, and to examine the relationship between latent class and psychological and behavioral problems. Data from 188 participants were analyzed. First, the latent class of adults’ non-suicidal self-injurious behaviors was identified as mild, moderate, or severe. Further analysis revealed that risk factors affecting self-injury latent class type included the experience of child abuse, lower levels of pain tolerance, younger age, and exposure to self-injury within interpersonal relationships. Levels of depression were highest in the severe class followed by the moderate and then mild classes. The severe class also displayed higher levels of anxiety and binge eating than the moderate and mild classes. The implications and limitations of these results are discussed.

현재 전승되는 조선시대 궁중음악 중 향악을 대표하는 악곡인 <수제천(壽齊天)>은 백제의 옛 가요로 알려진 <정읍>과의 관련성이 주목을 받고 있다. <수제천>의 본래 이름인 <정읍>은 오늘날의 전라북도 정읍시와 이름이 같아서, <수제천>이 본래 정읍지방의 옛 음악인지도 관심의 대상이다. 그동안 학계에서는 이와 관련한 많은 논의가 있었으나, 아직도 합의를 보이지는 않고 있다. 또한 조선 초기의 악보와 악서에 실린 <정읍>과 현재의 <수제천>이 같은 계통의 악곡인지도 여전히 논란의 대상이다, 그리고 최근에는 수제천의 악조와 관련한 이견도 제시되고 있어 이들 논의에 대한 재검토가 필요하다. 이 글에서는 위에서 제시된 여러 문제를 검토한 결과 다음과 같은 결론을 얻을 수 있었다. 1. <정읍>은 백제시대 정읍지방의 민요였으며, 한 행상인의 아내가 익숙한 민요 가락에 자신의 심경을 얹어 표현한 것으로 추정된다. 이후 <정읍>은 고려시대에 궁중무용의 반주곡으로 수용되었으며, 현재는 <수제천>이라는 기악곡으로 전승되고 있다. 2. 조선 전기 악보와 악서에 수록된 <정읍>은 3장으로 구성된 악곡으로, 2장은 1장의 반복이며, 3장은 1․2장을 4도 높게 변주한 것이다. 3. 현행 <수제천>의 악조는 1․2장이 남려계면조, 3장 3~6장단은 태주계면조이며, 3장 7장단에서 다시 남려계면조로 변조하고, 4장 제2장단에서 임종평조로 변조하여 악곡을 마친다. 이처럼 악조를 바꾸는 과정에서 3장 1․2장단과 4장의 첫 장단은 변조과정에서 교량역할을 하는 부분이다. 4. 조선 전기의 <정읍>과 현행 <수제천>은 악곡 형식구조의 공통점과 장단구조의 유사성, 그리고 ‘빗가락정읍’이라는 별칭으로 인하여 동일계 악곡으로 확인된다. 다만 500여년 이상의 시차를 지닌 두 악곡에서 발견되는 상이점은 시대에 따른 다양한 변화의 흔적이 적층된 결과로 보인다. ‘한국 궁중음악의 백미(白眉)’로 일컬어지는 <수제천>의 존재와 그 전승 의미는 다음과 같다. 첫째, 백제시대 정읍지방의 민요인 <정읍>이 1,000여년 이상의 오랜 시간동안 다양하게 변모되면서 오늘날 한국의 전통음악을 대표하는 악곡의 하나로 전승되고 있다는 점. 둘째, 현전 궁중음악 중 5음음계 계면조의 악곡이 거의 없는 상태에서 남려계면조, 태주계면조 등 5음음계 계면조의 악곡이 남아 옛 궁중음악의 자취를 남기고 있다는 점. 셋째, <수제천>은 궁중음악으로서는 매우 드물게 악곡 도중에서 악조를 바꾸는 음악적인 다양성을 지닌 음악이다. 이는 <수제천>이 단순한 의례주악용 음악이나 무용반주의 기능에 머무르지 않고, 변조를 통하여 다양한 음악적 미감(美感)을 추구한 음악임을 보여준다. 넷째, <수제천>은 민요가 궁중음악으로 수용되어 국가적인 의례에서 연주된 음악이다. 이처럼 민요가 궁중음악으로 수용되었다는 사실은 백성을 국가의 근본으로 여기는 ‘예악사상’ 구현의 상징적인 의미를 드러낸다.

“Sujecheon” is a representative hyangak (Korean indigenous music) and the court music currently transmitted from the Joseon period. Its relation to “Jeongeup”, known as a traditional popular song of Baekje Kingdom, has attracted academic consideration. “Jeongeup”, an original name of “Sujecheon”, is also a name of a village located in Jeollabuk-do province and this generates a question whether “Sujecheon” used to be a local song of the Jeongeup region. Although there has been much discussion about this issue in related academic fields, scholars have not yet reached a conclusive agreement. A comparative study has been still going on to decide whether the “Jeongeup” notation in Daeakhubo, the compilation of notation dated early Joseon period, and present “Sujecheon” are the same music. In addition, a new perspective was proposed a few years ago about the akjo (music structure, the modes and scales) of <Sujecheon>. Appropriate research is necessary for such issues. This paper has carried out a study to review all related problems and reached following conclusions. “Jeongeup” was a folk song of Jeongeup region during the period of Baekje Kingdom. It is traced that a wife of a peddler expressed her emotions with lyrics to sing upon an existing melody of a folk song familiar to her. “Jeongeup” was later adopted as an accompanying music for court dance during the Goryoe period and is now transmitted as an instrumental piece titled “Sujecheon”. Both “Jeongeup” score listed in the notation book of early Joseon period and present “Sujecheon” share similarity in their music forms and Jangdan structure. Current “Sujecheon” is also called “Bitgarak Jeongeup”. Based on these points, it is quite possible to conclude they should be the same piece. Current “Sujecheon” is in gyemyeonjomode, transposed to higher key in the third section and finally modulated to pyeongjomode in the fourth section, to end. The musical and cultural significations of “Sujecheon” as the best representative piece of all Korean traditional court music can be described as below. First, a folk song with the same name in a village named “Jeongeup” in Baekje Kingdom period has long been transmitted for about 1,000 years reshaping itself in many different ways and survived as one of the representative pieces of Korean traditional music. Second, even though it is hard to find pentatonic gyemyeonjo in the existing traditional court music, “Sujecheon” is valuable that it still holds the characteristics of traditional court music in its pentatonic scale in Gyemyeonjomode, including A Gyemyeonjo, F Gyemyeonjo. Third, “Sujecheon” is unique especially in that it adds musical diversity to itself through modulation in the middle. This proves that “Sujecheon” is not merely a functional music for ritual or dance accompaniment but carries aesthetic value music-wise, specially through modulating itself. Fourth, originally a folk song, “Sujecheon” was later adopted into court music and then played for state rites. Incorporating a folk song into court music reflects the symbolic meaning that “the ideology of the rite and music (Yeak ideology)’ is realized on the belief that people are the basis of nation.

In order to prevent emotional and behavioral problems and increase life satisfaction in early adolescents, this study classified mothers’ parenting attitudes perceived by early adolescents as latent classes, and verified emotional and behavioral problems and life satisfaction of early adolescents according to the classified latent classes. For this study, 2019 year’s data from the Korean Children and Youth Panel Survey(KCYPS) of the National Youth Policy Institute were used. A total of 4,368 cases(2,184 fifth grade in elementary school and 2,184 mothers) were statistically analyzed using latent profile analysis, multinominal logistic regression, and Wald’s association test. As results of the study, the latent classes for the mothers' parenting attitude were divided into three classes. As factors affecting the latent class, conversation time with mother in children, mother’s health, self-esteem and happiness in mothers significantly influenced to three latent classes. Also, there were significant differences in emotional and behavioral problems and life satisfaction according to the latent classes of mothers’ parenting attitudes. Emotional and behavioral problems were low and life satisfaction was high in the 'positive parenting attitude dominant group'. This study is expected to be useful in parent education and counseling sites.

본 연구는 현재 시행되고 있는 노인장기요양보험제도의 등급판정체계를 통합적, 실증적으로 살펴봄으로써 그 문제점과 적용상의 쟁점을 고찰하기 위한 것으로, 연구방법은 문헌연구와 등급판정 관련 전문가 대상 FGI를 실시하였다. 연구결과, 등급판정기준의 제한성이나 엄격성 문제와 인지적, 사회적 영역에 대한 과소평가 경향이 나타나 이에 대한 개선책이 필요한 것으로 나타났다. 그리고 평가내용과 관련해 정서적 영역을 측정하는 부분이 행동적 영역 안에 포함되어 전문적 평가가 이루어지지 못하는 제한이 있는 것으로 나타났고 대상자의 등급결정에 고려되어야 하는 중요한 요소들이 간과되고 있음이 보고되었다. 또한, 등급판정인력의 전문성과 의사소견서의 신뢰성 문제, 1차 방문조사원의 과다업무로 인한 질적 하락 문제가 있는 것으로 나타나 인력충원 및 이들에 대한 정기적인 재교육과 이들을 평가할 수 있는 시스템 마련이 필요한 것으로 나타났다. 등급판정절차와 관련해 노인장기요양서비스 공급구조의 왜곡된 시장화로 인한 편법과 불법현상 문제가 있는 것으로 나타났으며, 그 외 동거가족 수발제도에 대한 문제점이 상대적으로 높은 것으로 나타나 동거가족 수발의 현황 및 문제점에 대한 정부 차원의 체계적인 연구와 실태조사가 필요한 것으로 나타났다.

The purpose of this study is to analyze the main issues and problems of the the classification system of Long-Term Care Insurance(LTCI), which has been revealed many problems since it was launched in July 2008 and then to find the improvements. The research used the literature survey which analyzes 2nd-hand materials studied by other people, and Focus Group Interview was executed from 18 classification professionals in the area of classification system of LTCI. This research found the problems of the classification system of LTCI, which was about classification criteria, classification contents, classification committee, visiting investigator and classification procedures. Based on these findings, the study puts forward some suggestion regarding how to improve the classification system of LTCI. The LTCI organization requires more worker and higher professionality and efforts to resolve problems should be made for the efficient implementation and safe arrival of the classification system of LTCI.

본 연구는 초등학교 4학년 시기와 중학교 1학년 시기 정서행동문제의 전이 양상을 파악하고 각 시점의 잠재집단 분류에 영향을 미치는 예측요인을 분석하는 데 목적이 있다. 이를 위해 한국아동․청소년패널조사 2018의 초등학교 4학년 중 1차년도와 4차년도 조사에 참여한 2,154명의 자료를 분석에 활용하였으며, 잠재프로파일분석과 잠재전이분석을 하였다. 연구결과는 다음과 같다. 첫째, 초등학교 4학년 시기 정서행동문제 잠재집단은 4개로 분류되었으며, 각각은 특성에 따라 저수준, 중수준, 고수준, 우울 우세 집단으로 명명되었다. 둘째, 중학교 1학년 시기 정서행동문제 잠재집단은 5개로 분류되었으며, 저수준과 중수준, 고수준, 우울 우세, 외현화 우세, 위축 우세 집단으로 명명되었다. 셋째. 초등학교 4학년 시기 정서행동문제 수준이 높은 집단에 속할수록 중학교 1학년 시기 문제행동이 높은 집단에 속할 확률이 높았으며, 초등학교 4학년 우울 우세 집단의 경우 중학교 1학년 시기에도 같은 집단으로 전이될 확률이 높은 것으로 나타났다. 넷째. 초등학교 4학년 시기 정서행동문제 잠재프로파일의 분류를 예측하는 요인으로 자아존중감과 친구관계, 교사와 관계는 모두 유의한 영향을 미치는 것으로 나타났으며, 잠재집단에 따라 영향력의 양상이 상이한 것으로 나타났다. 다섯째, 중학교 1학년 시기 정서행동문제 잠재프로파일의 분류에는 자아존중감과 친구관계는 유의한 영향을 미쳤으나 교사와 관계는 유의한 영향을 미치지 않는 것으로 나타났다. 본 연구의 결과를 토대로 청소년의 정서행동문제에 대한 개입방안의 시사점 및 후속연구에 대하여 제언하였다.

This study examines the transition patterns of emotional and behavioral problems from the fourth grade of elementary school to the first grade of middle school and identifies key predictive factors influencing latent group classification at each time point. Using data from 2,154 students in the fourth-grade elementary school panel of the 2018 Korean Children and Youth Panel Survey (KCYPS 2018), latent profile analysis (LPA) and latent transition analysis (LTA) were conducted. The results revealed four latent groups in the fourth grade (low-level, moderate-level, high-level, and depression-dominant) and five in the first grade of middle school (low-level, moderate-level, high-level, depression-dominant, externalizing-dominant, and withdrawal-dominant). Students in higher-risk groups in elementary school were more likely to transition into higher-risk groups in middle school, with those in the depression-dominant group showing strong stability across time. Additionally, self-esteem, peer relationships, and teacher-student relationships significantly predicted the classification of latent profiles in elementary school, though their influence varied by group. In middle school, self-esteem and peer relationships remained significant predictors, whereas teacher-student relationships did not have a significant effect. Based on these findings, implications for interventions targeting adolescents’ emotional and behavioral problems and suggestions for future research are discussed.

With the wealth of image data that is now becoming increasingly accessible through the advent of the world wide web and the proliferation of cheap, high quality digital cameras it is becoming ever more desirable to be able to automatically classify images into appropriate categories such that intelligent agents and other such intelligent software might make better informed decisions regarding them without a need for excessive human intervention. However, as with most Artificial Intelligence (A.I.) methods it is seen as necessary to take small steps towards your goal. With this in mind a method is proposed here to represent localised features using disjoint sub-images taken from several datasets of retinal images for their eventual use in an incremental learning system. A tile-based localised adaptive threshold selection method was taken for vessel segmentation based on separate colour components. Arteriole-venous differentiation was made possible by using the composite of these components and high quality fundal images. Performance was evaluated on the DRIVE and STARE datasets achieving average specificity of 0.9379 and sensitivity of 0.5924.

Mine Blast Algorithm (MBA) is newly developed metaheuristic technique. It has outperformed Genetic Algorithm (GA), Particle Swarm Optimization (PSO) and their variants when solving various engineering optimization problems. MBA has been improved by IMBA, which is modified in this paper to accelerate its convergence speed furthermore. The proposed variant, so called Accelerated MBA (AMBA), replaces the previous best solution with the available candidate solution in IMBA. ANFIS accuracy depends on the parameters it is trained with. Keeping in view the drawbacks of gradients based learning of ANFIS using gradient descent and least square methods in two-pass learning algorithm, many have trained ANFIS using metaheuristic algorithms. In this paper, for getting high performance, the parameters of ANFIS are trained by the proposed AMBA. The experimental results of real-world benchmark problems reveal that AMBA can be used as an efficient optimization technique. Moreover, the results also indicate that AMBA converges earlier than its other counterparts MBA and IMBA.

The standard back propagation algorithm for training artificial neural networks utilizes two terms, a learning rate and a momentum factor. The major limitations of this standard algorithm are the existence of temporary, local minima resulting from the saturation behaviour of the activation function, and the slow rates of convergence. Previous research demonstrated that in ‘feed forward’ algorithm, the slope of the activation function is directly influenced by a parameter referred to as ‘gain’. This research proposed an algorithm for improving the performance of the back propagation algorithm by introducing the adaptive gain of the activation function. The efficiency of the proposed algorithm is compared with conventional Gradient Descent Method and verified by means of simulation on four classification problems. In learning the patterns, the simulations result demonstrate that the proposed method converged faster on Wisconsin breast cancer and diabetes classification problem with an improvement ratio of nearly 2.8 and 1.2, 65% better on thyroid data sets and 97% success on IRIS classification problem. The results clearly show that the proposed algorithm significantly improves the learning speed of the conventional back-propagation algorithm.

The back propagation (BP) algorithm is a very popular learning approach in multilayer feedforward networks. However, the most serious problems associated with the BP are local minima problem and slow convergence speeds. Over the years, many improvements and modifications of the BP learning algorithm have been reported. In this research, we propose a new modified BP learning algorithm by introducing adaptive gain together with adaptive momentum and adaptive learning rate into weight update process. By computer simulations, we demonstrate that the proposed algorithm can give a better convergence rate and can find a good solution in early time compare to the conventional BP. We use four common benchmark classification problems to illustrate the improvement in convergence time.

In order to obtain more value added products, a product quality control is essentially required Many studies show that quality of agriculture products may be reduced from many sources. In this paper I express Technological Strategies uses Sony cyber-shot DSC W350 camera to captured symptoms of Brassica sp. diseased leaf images and sort out the diseases using support vector machine. There is a need to classify and regrow them for the use of future generations. So it is very indispensable to develop a spontaneous system to identify species correctly. This paper proposes an algorithm for recognition of diseased leaf and show the name of disease through image processing and classification technique using Support Vector Machine (SVM) classifier. The K-Mean Clustering method is used for Segmentation and Gray-level co-occurrence matrix (GLCM) is used for the feature extraction in leaves. The work is carried out using 2 species of leaves in MATLAB environment.

2017년 국립대학자원관리시스템(KORUS)이 개통되면서 국립대학의 기록관리체계도 단 위과제 중심의 기록관리기준표 분류체계로의 전환점을 맞이하였다. 국립대학은 표준화된 방식으로 운영되는 행정시스템을 기반으로 업무처리 과정을 단위과제카드를 중심으로 기록물을 생산⋅관리하게 되었다. 특히, 협업과 정보공유를 위해 대학의 공통 업무에 일관된 관리기준을 부여하는 공통 단위과제를 개발하여 운영하고 있다. 본고에서는 현재 국립대학에서 사용하는 기록물 분류체계의 운영현황과 운영과정에서 나타나는 문제점을 공통 단위과제를 중심으로 분석하였다. 국립대학 기록물의 특성과 분류체계 개발 경과, 운영현황을 분석하여 기록관리기준표가 기록관리 도구로의 역할을 이행하고 있는지를 확인하고 기록분류 도구이자 평가와 검색도구로 기능할 수 있도록 개선방안을 제안하였다.

With the Korean University Resource United System (KORUS) opening in 2017, the national university’s record management system has been turning into a unit task-oriented records management reference table. Based on the administration system operated in a standardized manner, national universities can produce and manage university records based on business transaction cards. In the process, the standard business transactions for universities’ common tasks were developed and operated for collaboration and sharing information. This study aimed to analyze the national university classification system’s current operating status and problems, focusing on common unit tasks. To this end, the characteristics of university records, the development process of the national university classification system, and its operating status were analyzed. It also investigated whether the records management reference table is suitable for use as a record management tool. In conclusion, this study suggested improvements on the national university records management reference table, which can function as a record classification and appraisal tool.

We describe a new deep learning model - Convolutional eXtreme Gradient Boosting (ConvXGB) for classification problems based on convolutional neural nets and Chen et al.'s XGBoost. As well as image data, ConvXGB also supports the general classification problems, with a data preprocessing module. ConvXGB consists of several stacked convolutional layers to learn the features of the input and is able to learn features automatically, followed by XGBoost in the last layer for predicting the class labels. The ConvXGB model is simplified by reducing the number of parameters under appropriate conditions, since it is not necessary re-adjust the weight values in a back propagation cycle. Experiments on several data sets from UCL Repository, including images and general data sets, showed that our model handled the classification problems, for all the tested data sets, slightly better than CNN and XGBoost alone and was sometimes significantly better.

The usual method of classification for metal poor stars is based on the normal standard stars. In this study, we show that among the sample of stars classified by this method, a systematic bias in the observed classes of metal weakness is found and, also that this method is not appropriate for classification of metal poor stars, by showing that the spectral line dependences on the temperature and pressure in the extreme metal poor stars are different from those in the normal standard stars. Therefore, we suggest that the 3-dimensional classification system, like 2-dimensional MK system, is necessary for an accurate classification of metal poor stars.

최근 알파고의 등장으로 딥러닝 기술에 대한 관심이 고조되고 있다. 딥러닝은 향후 미래의 핵심 기술이 되어 일상생활의 많은 부분을 개선할 것이라는 기대를 받고 있지만, 주요한 성과들이 이미지 인식과 자연어처리 등에 국한되어 있고 전통적인 비즈니스 애널리틱스 문제에의 활용은 미비한 실정이다. 실제로 딥러닝 기술은 Convolutional Neural Network(CNN), Recurrent Neural Network(RNN), Deep Boltzmann Machine (DBM) 등 알고리즘들의 선택, Dropout 기법의 활용여부, 활성 함수의 선정 등 다양한 네트워크 설계 이슈들을 가지고 있다. 따라서 비즈니스 문제에서의 딥러닝 알고리즘 활용은 아직 탐구가 필요한 영역으로 남아있으며, 특히 딥러닝을 현실에 적용했을 때 발생할 수 있는 여러 가지 문제들은 미지수이다. 이에 따라 본 연구에서는 다이렉트 마케팅 응답모델, 고객이탈분석, 대출 위험 분석 등의 주요한 분류 문제인 이진분류에 딥러닝을 적용할 수 있을 것인지 그 가능성을 실험을 통해 확인하였다. 실험에는 어느 포르투갈 은행의 텔레마케팅 응답여부에 대한 데이터 집합을 사용하였으며, 전통적인 인공신경망인 Multi-Layer Perceptron, 딥러닝 알고리즘인 CNN과 RNN을 변형한 Long Short-Term Memory, 딥러닝 모형에 많이 활용되는 Dropout 기법 등을 이진 분류 문제에 활용했을 때의 성능을 비교하였다. 실험을 수행한 결과 CNN 알고리즘은 비즈니스 데이터의 이진분류 문제에서도 MLP 모형에 비해 향상된 성능을 보였다. 또한 MLP와 CNN 모두 Dropout을 적용한 모형이 적용하지 않은 모형보다 더 좋은 분류 성능을 보여줌에 따라, Dropout을 적용한 CNN 알고리즘이 이진분류 문제에도 활용될 수 있는 가능성을 확인하였다.

Recently, AlphaGo which is Bakuk (Go) artificial intelligence program by Google DeepMind, had a huge victory against Lee Sedol. Many people thought that machines would not be able to win a man in Go games because the number of paths to make a one move is more than the number of atoms in the universe unlike chess, but the result was the opposite to what people predicted. After the match, artificial intelligence technology was focused as a core technology of the fourth industrial revolution and attracted attentions from various application domains. Especially, deep learning technique have been attracted as a core artificial intelligence technology used in the AlphaGo algorithm. The deep learning technique is already being applied to many problems. Especially, it shows good performance in image recognition field. In addition, it shows good performance in high dimensional data area such as voice, image and natural language, which was difficult to get good performance using existing machine learning techniques. However, in contrast, it is difficult to find deep leaning researches on traditional business data and structured data analysis. In this study, we tried to find out whether the deep learning techniques have been studied so far can be used not only for the recognition of high dimensional data but also for the binary classification problem of traditional business data analysis such as customer churn analysis, marketing response prediction, and default prediction. And we compare the performance of the deep learning techniques with that of traditional artificial neural network models. The experimental data in the paper is the telemarketing response data of a bank in Portugal. It has input variables such as age, occupation, loan status, and the number of previous telemarketing and has a binary target variable that records whether the customer intends to open an account or not. In this study, to evaluate the possibility of utilization of deep learning algorithms and techniques in binary classification problem, we compared the performance of various models using CNN, LSTM algorithm and dropout, which are widely used algorithms and techniques in deep learning, with that of MLP models which is a traditional artificial neural network model. However, since all the network design alternatives can not be tested due to the nature of the artificial neural network, the experiment was conducted based on restricted settings on the number of hidden layers, the number of neurons in the hidden layer, the number of output data (filters), and the application conditions of the dropout technique. The F1 Score was used to evaluate the performance of models to show how well the models work to classify the interesting class instead of the overall accuracy. The detail methods for applying each deep learning technique in the experiment is as follows. The CNN algorithm is a method that reads adjacent values from a specific value and recognizes the features, but it does not matter how close the distance of each business data field is because each field is usually independent. In this experiment, we set the filter size of the CNN algorithm as the number of fields to learn the whole characteristics of the data at once, and added a hidden layer to make decision based on the additional features. For the model having two LSTM layers, the input direction of the second layer is put in reversed position with first layer in order to reduce the influence from the position of each field. In the case of the dropout technique, we set the neurons to disappear with a probability of 0.5 for each hidden layer. The experimental results show that the predicted model with the highest F1 score was the CNN model using the dropout technique, and the next best model was the MLP model with two hidden layers using the dropout technique. In this study, we were able to get some findings as the experiment had proceeded. First, models using dropout techniques have a slightly more conse

이진 분류기로서 지지벡터기계는 다양한 응용을 통해 이진 분류 문제에서 기존의 패턴 분류기들보다 우수한 성능을 보였다. 지지벡터기계의 바탕이 되는 최대 마진 분류 이론을 다중 분류 문제에 확장은 어려움이 있다. 이 논문에서는 다중 분류 문제를 위한 지지벡터기계의 학습 전략을 논의하였으며 성능 비교를 수행하였다. 학습 데이터의 분배 전략에 따라 지지벡터기계는 고유의 이진 분류 특징을 수정하지 않고 다중분류 문제에 쉴게 적용될 수 있다. 다양한 벤치마킹 데이터에 대해 선택된 학습 전략, 커널함수, 학습 소요시간 등에 따라 성능비교가 수행되었고 오류역전파 학습의 신경망의 테스트 결과와 비교되었다. 신경망 모델과 비교 실험에서 지지벡터기계는 일반적인 다중 분류 문제에 응용성과 효과가 있음을 보였다.

The support vector machine, as a binary classifier, is known to surpass the other classifiers only in binary classification problems through the various experiments. Even though its theory is based on the maximal margin classifier, the support vector machine approach cannot be easily extended to the multi-classification problems. In this paper, we review the extension techniques of the support vector machine toward the multi-classification and do the performance comparison. Depending on the data decomposition of the training data, the support vector machine is easily adapted for a multi-classification problem without modifying the intrinsic characteristics of the binary classifier. The performance is evaluated on a collection of the benchmark data sets and compared according to the selected teaming strategies, the training time, and the results of the neural network with the backpropagation teaming. The experiments suggest that the support vector machine is applicable and effective in the general multi-class classification problems when compared to the results of the neural network.

최근 컴퓨터를 이용한 다양한 인식 문제를 해결하기 위해 딥 러닝을 적용하는 사례가 늘어나고 있다. 딥 러닝은 학습에 필요한 요소를 학습데이터를 통해 스스로 도출해내기 때문에, 수작업(hand-craft)을 통해 특징을 도출하던 기존의 기계학습 방법보다 더 많은 장점을 갖는다. 행동인식을 위한 기존의 심층 신경망은 비디오 데이터를 일정 프레임의 이미지로 분할한 후, 분할된 각 이미지 사이의 시간적 연계성 분석을 통해 행동을 분류한다. 그러나 이러한 신경망은 소규모 행동 클래스를 갖는 분류 문제에서 학습 데이터의 부족 문제 및 과적합(overfitting) 문제로 인해 이를 실제 문제에 적용하기 어려운 경우가 많다. 이에 본 논문에서는 5가지의 소규모 행동 클래스를 정의하고, 기존 행동 인식 신경망의 최적화를 통해 이를 분류하였다. 700개의 비디오데이터를 통해 행동 데이터베이스를 구성하였고, 약 74.00%의 분류 정확도를 얻을 수 있었다.

Recently, Deep learning has been used successfully to solve many recognition problems. It has many advantages over existing machine learning methods that extract feature points through hand-crafting. Deep neural networks for human activity recognition split video data into frame images, and then classify activities by analysing the connectivity of frame images according to the time. But it is difficult to apply to actual problems which has small-scale activity classes. Because this situations has a problem of overfitting and insufficient training data. In this paper, we defined 5 type of small-scale human activities, and classified them. We construct video database using 700 video clips, and obtained a classifying accuracy of 74.00%.