• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.221-226
• /
• 2021

우리는 범주 불균형 분류 문제를 해결하기 위해 학습 과정 중 범주 크기 기반 배치 샘플링 방법 전환을 위한 스케줄링 방법을 제안한다. 범주별 샘플링 확률로 범주 크기의 역수(LWRS-Reciporcal)와 범주 비율의 반수(LWRS-Ratio)를 적용하여 각각 실험을 진행하였고, LWRS-Reciporcal 방법이 F1 성능 개선에 더 효과적인 것을 확인하였다. 더하여 고정된 샘플링 확률값으로 인해 발생할 수 있는 또 다른 편향 문제를 완화하기 위해 학습 과정 중 샘플링 방법을 전환하는 스케줄링 방법을 설계하였다. 결과적으로 검증 성능의 갱신 유무로 샘플링 방법을 전환하였을 때 naver shopping 데이터셋과 KLUE-TC에 대하여 f1 score와 accuracy의 성능 합이 베이스라인보다 각각 0.7%, 0.8% 향상된 가장 이상적인 성능을 보임을 확인하였다.

• KIPS Transactions on Computer and Communication Systems
• /
• v.9 no.4
• /
• pp.77-82
• /
• 2020

Wireless sensor networks (WSNs) suffer from not only a short lifetime due to limited energy but also an energy imbalance between nodes close to the sink and others. In order to fundamentally solve the short lifetime, recent studies utilize the environmental energy such as solar power. Additionally, WSNs using mobile sinks are being studied to address the energy imbalance problem. This paper proposes an improved CTP (Collection Tree Protocol) scheme which uses these two approaches simultaneously. Basically, it is based on a CTP scheme which is a very popular data collection strategy designed for the typical battery-based WSNs with a fixed sink. Therefore, we tailored it for solar-powered WSNs with a mobile sink. Performance verification confirms that our scheme reduces the number of blackout nodes significantly compared to the typical CTP, thus increases the amount of data collected by the sink.

• Information Systems Review
• /
• v.9 no.1
• /
• pp.45-65
• /
• 2007

In medical industry, health insurance bill audit is unique and essential process in general hospitals. The health insurance bill audit process is very important because not only for hospital's profit but also hospital's reputation. Particularly, at the large general hospitals many related workers including analysts, nurses, and etc. have engaged in the health insurance bill audit process. This paper introduces a case of health insurance bill audit for finding reducible health insurance bill cases using decision tree induction techniques at a large general hospital in Korea. When supervised learning methods had been tried to be applied, one of major problems was data imbalance problem in the health insurance bill audit data. In other words, there were many normal(passing) cases and relatively small number of reduction cases in a bill audit dataset. To resolve the problem, in this study, well-known methods for imbalanced data sets including over sampling of rare cases, under sampling of major cases, and adjusting the misclassification cost are combined in several ways to find appropriate decision trees that satisfy required conditions in health insurance bill audit situation.

• Journal of the Korea Society of Computer and Information
• /
• v.12 no.3
• /
• pp.51-58
• /
• 2007

The shared-nothing multiprocessor architecture has advantages in scalability, this architecture has been adopted in many multiprocessor database system. But, if the data are not uniformly distributed across the processors, load will be unbalanced. Therefore, the whole system performance will deteriorate. This is the data skew problem, which usually occurs in processing parallel hash join. Balancing the load before performing join will resolve this problem efficiently and the whole system performance can be improved. In this paper, we will present an algorithm using merit of very large memory to reduce disk access overhead in performing load balancing and to efficiently solve the data skew problem. Also, we will present analytical model of our new algorithm and present the result of some performance study we made comparing our algorithm with the other algorithms in handling data skew.

• Smart Media Journal
• /
• v.13 no.2
• /
• pp.9-15
• /
• 2024

In this paper, a Diffusion Multi-step Classifier (DMC) is proposed to address the imbalance issue in credit prediction. DMC utilizes a Diffusion Model to generate continuous numerical data from credit prediction data and creates categorical data through a Multi-step Classifier. Compared to other algorithms generating synthetic data, DMC produces data with a distribution more similar to real data. Using DMC, data that closely resemble actual data can be generated, outperforming other algorithms for data generation. When experiments were conducted using the generated data, the probability of predicting delinquencies increased by over 20%, and overall predictive accuracy improved by approximately 4%. These research findings are anticipated to significantly contribute to reducing delinquency rates and increasing profits when applied in actual financial institutions.

• Information Systems Review
• /
• v.18 no.2
• /
• pp.173-198
• /
• 2016

Financial distress can damage stakeholders and even lead to significant social costs. Thus, financial distress prediction is an important issue in macroeconomics. However, most existing studies on building a financial distress prediction model have only considered idiosyncratic risk factors without considering systematic risk factors. In this study, we propose a prediction model that considers both the idiosyncratic risk based on a financial ratio and the systematic risk based on a business cycle. Ultimately, we build several IT artifacts associated with financial ratio and add them to the idiosyncratic risk factors as well as address the imbalanced data problem by using an oversampling technique and synthetic minority oversampling technique (SMOTE) to ensure good performance. When considering systematic risk, our study ensures that each data set consists of both financially distressed companies and financially sound companies in each business cycle phase. We conducted several experiments that change the initial imbalanced sample ratio between the two company groups into a 1:1 sample ratio using SMOTE and compared the prediction results from the individual data set. We also predicted data sets from the subsequent business cycle phase as a test set through a built prediction model that used business contraction phase data sets, and then we compared previous prediction performance and subsequent prediction performance. Thus, our findings can provide insights into making rational decisions for stakeholders that are experiencing an economic crisis.

• KIPS Transactions on Computer and Communication Systems
• /
• v.11 no.12
• /
• pp.445-452
• /
• 2022

Class distribution of unbalanced data is an important part of the digital world and is a significant part of cybersecurity. Abnormal activity of unbalanced data should be found and problems solved. Although a system capable of tracking patterns in all transactions is needed, machine learning with disproportionate data, which typically has abnormal patterns, can ignore and degrade performance for minority layers, and predictive models can be inaccurately biased. In this paper, we predict target variables and improve accuracy by combining estimates using Synthetic Minority Oversampling Technique (SMOTE) and Light GBM algorithms as an approach to address unbalanced datasets. Experimental results were compared with logistic regression, decision tree, KNN, Random Forest, and XGBoost algorithms. The performance was similar in accuracy and reproduction rate, but in precision, two algorithms performed at Random Forest 80.76% and Light GBM 97.16%, and in F1-score, Random Forest 84.67% and Light GBM 91.96%. As a result of this experiment, it was confirmed that Light GBM's performance was similar without deviation or improved by up to 16% compared to five algorithms.

• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.507-511
• /
• 2020

오픈 도메인 대화에서 텍스트에 나타난 태도나 성향과 같은 화자의 주관적인 감정정보를 분석하는 것은 사용자들에게서 풍부한 응답을 이끌어 내고 동시에 제공하는 목적으로 사용될 수 있다. 하지만 한국어 감성분석에서 기존의 대부분의 연구들은 긍정과 부정 두개의 클래스 분류만을 다루고 있고 이는 현실 화자의 감정 정보를 정확하게 분석하기에는 어려움이 있다. 또한 최근에 오픈한 다중클래스로된 한국어 대화 감성분석 데이터셋은 중립 클래스가 전체 데이터셋의 절반을 차지하고 일부 클래스는 사용하기에 매우 적은, 다시 말해 클래스 간의 데이터 불균형 문제가 있어 다루기 굉장히 까다롭다. 이 논문에서 우리는 일곱개의 클래스가 존재하는 한국어 대화에서 세션들을 효율적으로 분류하는 기법들에 대해 논의한다. 우리는 극심한 클래스 불균형에도 불구하고 76.56 micro F1을 기록하였다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.91-93
• /
• 2020

일상의 데이터를 감성 분석에 사용할 때, 데이터 개수가 부족하거나 불균형한 문제가 발생한다. 이에 대해, 본 논문은 데이터 수집 카테고리를 확장하는 방법으로 기존의 문제들을 해결한다. 나아가 확장된 카테고리로 수집한 데이터를 이용해 감성 분석 모델을 만들고, 해당 카테고리에 속하는 타겟 제품에 대한 감성 분석을 시도한다.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.251-256
• /
• 2021

개체명 인식은 정보 추출의 하위 작업으로, 문서에서 개체명에 해당하는 단어를 찾아 알맞은 개체명을 분류하는 자연어처리 기술이다. 질의 응답, 관계 추출 등과 같은 자연어처리 작업에 대한 관심이 높아짐에 따라 세부 분류 개체명 인식에 대한 수요가 증가했다. 그러나 기존 개체명 인식 성능에 비해 세부 분류 개체명 인식의 성능이 낮다. 이러한 성능 차이의 원인은 세부 분류 개체명 데이터가 불균형하기 때문이다. 본 논문에서는 이러한 데이터 불균형 문제를 해결하기 위해 대분류 개체명 정보를 활용하여 세부 분류 개체명 인식을 수행하는 방법과 대분류 개체명 인식의 오류 전파를 완화하기 위한 2단계 학습 방법을 제안한다. 또한 레이블 주의집중 네트워크 기반의 구조에서 레이블의 공통 요소를 공유하여 세부 분류 개체명 인식에 효과적인 레이블 임베딩 구성 방법을 제안한다.