Earticle

A development in method of evaluating the measurement precisions using approximate F tests and variance components from expected mean square (EMS) is investigated. The research proposes three-factor mixed measurement models with the fixed and random factors. Unrestricted and unconstrained design work was rarely studied, while restricted and constrained designs have been significantly discussed. The unrestricted and unconstrained designs assume to be an independence of interaction. The proposed evaluation method about the measurement precisions can be extended to four-factor random measurement model or mixed measurement model. The study also presents the three evaluation indexes of precisions such as R&RTR (Reproducibility & Repeatability-To-Total Precision Ratio), PTR (Precision-To-Tolerance Ratio), and SNR (Signal-To-Noise Ratio). Numerical examples are proposed to evaluate the approximate F tests with Satterthwaite degrees of freedom and three indexes using the measurement precisions from EMS.

Purpose: The goal of this study was to explore the current status and influencing factors of nurses'intention to report adverse nursing events, using a random forest model to rank and select these factors. Methods: In August 2024, a survey was conducted among nurses at a tertiary hospital in Shanghai usinga convenience sampling method. The instruments included the General Information Questionnaire, theIntention to Report Questionnaire, the Reporting of Clinical Adverse Effects Scale (RoCAES), and theReporting Barrier Questionnaire. A random forest model and stepwise multiple linear regression wereused to analyze the factors influencing nurses' intention to report nursing adverse events. Results: The median total score for the intention to report adverse events among 823 nurses was 14(range: 0 to 15). Lasso regression identified eight key factors influencing the intention to report adverseevents: professional title, whether the nurse had reported adverse events, age, punitive culture,reporting process, reporting significance, reporting purpose, and reporting environment. Stepwiseregression analysis revealed that younger nurses, those with no prior reporting experience, and thosewho perceived reporting as important had a higher intention to report adverse events. Additionally,nurses were more likely to report adverse events when the hospital's reporting process was simpler andthe culture was less punitive. Conclusions: Nurses' intention to report adverse events in nursing is moderate. Nursing managers shouldconsider these influencing factors holistically and implement targeted measures to enhance nurses'intention to report, thereby improving nursing quality and patient safety.

본 연구는 임의회귀모형을 이용하여 한우 거세우 체중에 대해서 유전모수 추정을 하고 이것을 단형질 개체모형의 결과와 비교해 보고자 실시하였다. 분석에 이용한 자료는 총 1,372두의 한우 거세우의 체중 자료로, 농협중앙회 가축개량사업소에서 실시한 한우 후대검정우의 기록이다. 이차의 임의회귀 모형에 적용한 결과 유전력이 총 800일령까지의 검정기간에 대해 0.17~0.30의 범위로 나타났다. 개체모형을 통해 얻은 유전력은 0.24~0.36의 범위로 나타났다. 측정일간의 영구환경효과의 상관은 검정일령이 늘어남에 따라 함께 증가하는 경향을 보였다. 반면, 측정일간의 유전상관의 경우 검정초기에는 0.30 정도의 약한 음의 상관을 보이지만, 검정이 이뤄짐에 따라 상관이 점차 증가하여 검정이 종료될 무렵이면 거의 고정되는 것으로 나타났다. 임의회귀모형과 개체모형의 결과를 비교해보면 두가지 모형 모두 비슷한 경향을 보여 큰 차이를 보이지 않았다. 따라서, 임의회귀모형을 한우에 대한 국가유전능력평가에 사용하는 것이 가능할 것으로 사료된다.

The study aimed to estimate genetic parameters of body weights in Hanwoo steers using random regression model and compare it with single trait animal model. A total of 1,372 Hanwoo steers that belonged to progeny testing program of the Hanwoo Genetic Improvement conducted at the Livestock Improvement Main Center of the National Agricultural Cooperative Federation (LIMC-NACF) in Rep. of Korea were used. Results of the random regression model fitting quadratic function revealed heritability values from 0.17 to 0.30 for the whole testing days up to 800 days. The results of the animal model showed estimated heritability values ranged from 0.24 to 0.36. Estimates of permanent environmental correlations tended to increase with increasing test in days. Unlike in the direct genetic correlation that at early stage the estimate was slightly negative it was 0.30 then increased to approach unity at later stage of test. Comparing the results between random regression model and the animal model showed not much differences and both followed similar pattern and therefore the use of random regression model for the national genetic evaluation of Hanwoo could be implemented.

Beliefs regarding the controllability of emotions are closely associated with emotional distress, such as depression and anxiety. Most previous studies have focused on the unidirectional prediction of emotional distress based on emotion controllability beliefs. However, it is equally plausible that emotion controllability beliefs and emotional distress influence each other in a bi- directional manner. Using a random intercept cross-lagged panel model (RI-CLPM), this study aimed to elucidate the direc- tionality of the relationship between emotion controllability beliefs and emotional distress, with a focus on depression and anxiety. A total of 393 participants reported their emotion beliefs, and symptoms of depression and anxiety at time 1. They were followed up at 5 weeks (time 2), and 10 weeks (time 3). At the within-person level, higher-than-average levels of a per- son’s depression and anxiety symptoms predicted increases in beliefs that depression and anxiety are uncontrollable. In addi- tion, we found a significant bidirectional relationship between anxiety controllability beliefs and anxiety symptoms. These findings highlight the need to expand our current understanding of emotion controllability beliefs to account for the recipro- cal relationships between these beliefs and emotional distress, especially in the context of anxiety.

Randomness exists in engineering. Tolerance, assemble-error, environment temperature and wear make the parameters of a mechanical system uncertain. So the behavior or response of the mechanical system is uncertain. In this paper, the uncertain parameters are treated as random variables. So if the probability distribution of a random parameter is known, the simulation of mechanical multibody dynamics can be made by Monte-Carlo method. Thus multibody dynamics simulation results can be obtained in statistics. A new concept called functional reliability is put forward in this paper, which can be defined as the probability of the dynamic parameters(such as position, orientation, velocity, acceleration etc.) of the key parts of a mechanical multibody system belong to their tolerance values. A flexible mechanical arm with random parameters is studied in this paper. The length, width, thickness and density of the flexible arm are treated as random variables and Gaussian distribution is used with given mean and variance. Computer code is developed based on the dynamic model and Monte-Carlo method to simulate the dynamic behavior of the flexible arm. At the same time the end effector's locating reliability is calculated with circular tolerance area. The theory and method presented in this paper are applicable on the dynamics modeling of general multibody systems.

Purpose: Nursing homes face a growing demand for high-quality palliative wound care. Nursing home staff require not only knowledge but also clinical judgment to address the complex needs of residents at the end of life. This study aimed to develop and evaluate a web-based palliative wound care education program using a clinical judgment model. Methods: An educational program was developed using the ADDIE model through needs analysis with 10 nursing home staff, program design and development, and pilot implementation and evaluation. Program effectiveness was evaluated in a quasi-experimental study with facility-level random allocation. Four nursing homes were randomly allocated to the experimental or control condition, and 52 nursing home staff were included in the final analysis (n=26 per group). The experimental group received clinical judgment model-based web education, whereas the control group received general web-based palliative wound care education without the clinical judgment model. Clinical judgment was assessed before and after the intervention using nonparametric tests. Results: Both groups showed significant improvements in clinical judgment. However, the experimental group demonstrated greater improvements than the control group in noticing (Z=-3.91, p<.001), interpreting (Z=-3.39, p=.001), responding (Z=-3.64, p<.001), reflecting (Z=-2.30, p=.021), and the total clinical judgment score (Z=-4.62, p<.001). Conclusion: The findings support the effectiveness of clinical judgment model-based education in improving clinical judgment in palliative wound care. This web-based approach supports continuing education and improves care quality in nursing homes.

토석류는 우리나라 주요 산지토사재해 중 하나로, 토석류로 운송된 다량의 토사가 하류 지역에 퇴적 되어 인명 및 재산피해를 일으키기 때문에 토석류로 인한 피해 범위를 예측하여 경계 피난을 위한 기초자료로 활용하는 것이 중요하다. 우리나라 산림청은 Random Walk Model(RWM)을 활용하여 토석류 피해예측지도를 구축하였다. 한편, 산지계류에 설치되는 사방댐은 토석류 유출을 차단하여 피해면적을 감소시키므로 효과적인 토석류 피해예측지도 활용을 위해서는 사방댐의 효과가 고려되 어야 한다. 이 연구에서는 강원도 소재 9개 사방댐 설치 유역에 대해, 사방댐의 효과를 고려한 RWM 토석류 유출 및 퇴적 모의를 통해 토석류 피해예측지도에 대한 사방댐 효과 적용 가능성을 검토하였 다. 분석결과 RWM을 통해 사방댐의 효과는 일정 부분 재현 가능하였으나, 입력 매개변수 및 모델 구성의 한계로 피해예측범위의 수정은 제한될 것으로 판단되었다. 향후 공간해상도의 개선 및 합리 적 붕괴토사량의 산정을 통해 사방댐의 효과를 반영한 효과적인 토석류 피해예측지도 구축이 가능 할 것으로 판단된다.

Debris flow transports lots of sediments downstream that cause damage to human lives and properties. The accurate estimation of the debris flow hazard zone is a critical factor of the successful early-warning and evacuation for reducing the debris flow risk. Korea Forest Service has developed a debris flow hazard map using Random Walk Model (RWM). Meanwhile, because check dams in forest watersheds capture discharged debris, the effect of check dams should be considered for mapping the debris flow hazard zone. This study analyzed the effect of check dams on debris flow discharge using RWM with the check dam module. As a result, RWM seems to simulate the sediment capture by check dams. However, the total deposit areas were not significantly different despite the effect of check dams because of input parameters and the flow direction calculation algorithm of RWM. Further studies on appropriate spatial resolution and initial sediment volume should be conducted to improve the debris flow hazard map.

In this study, AAHT was estimated using machine learning algorithms of ensemble techniques to improve the limitations of existing studies. Among the machine learning algorithms, a random forest algorithm was used, and the traffic volume estimation accuracy was analyzed as MAPE 20.7%. It was analyzed that the higher the traffic volume level, the higher the traffic volume estimation accuracy, and the traffic volume level of 3,000 vehicles/hour or more was analyzed to be 8% or less of MAPE. It was analyzed that the accuracy secured at the current level was very high compared to the existing model.

High resolution rainfall data at 1-hour or a finer scale are essential for reliable flood analysis and forecasting; nevertheless, many observations, forecasts, and climate projections are still given at coarse temporal resolutions. This study aims to evaluate a chaotic method for disaggregation of 6-hour rainfall data sets so as to apply operational 6-hour rainfall forecasts of the Korean Meteorological Association to flood models. We computed parameters of a state-of-the-art multiplicative random cascade model with two combinations of cascades, namely uniform splitting and diversion, using rainfall observations at Seoul station, and compared statistical performance. We additionally disaggregated 6-hour rainfall time series at 58 stations with the uniform splitting and evaluated temporal transferability of the parameters and changes in multifractal properties. Results showed that the uniform splitting outperformed the diversion in reproduction of observed statistics, and hence is better to be used for disaggregation of 6-hour rainfall forecasts. We also found that multifractal properties of rainfall observations has adequate temporal consistency with an indication of gradually increasing rainfall intensity across South Korea.

집단 비교 연구 시 측정동일성의 성립 여부는 집단 간 의미 있는 비교를 하기 위한 필수 요건으로 제시되고 있다. 이를 위해 일반적으로 다집단 확인적요인분석(MG CFA)이 널리 사용되어 왔으나, MG CFA는 비교집단이 소수일 경우에 적합한 것으로, 많은 집단을 비교하게 되는 국가 비교 연구에서는 그 한계가 제기된다. 따라서 본 연구에서는 10개 이상의 많은 수의 국가(혹은 집단)를 비교하기에 적절한 방법론인 다층 확인적요인분석(ML CFA)과 다층 요인혼합모형(ML FMM)을 이용한 분석 방법론을 기술하였다. ML CFA는 절편만을 임의효과로 추정하는 임의절편모형과 요인계수도 임의효과로 추정하는 임의절편 및 임의요인계수모형으로 구분하여 장단점을 기술하였다. 구체적으로 각 방법론에 대한 이론적 모형과 측정동일성 검정 절차를 제시하고, 기존의 MG CFA에 비해 지니는 이점 및 적용 시 유의해야 할 사항을 서술하였다. 또한 이러한 방법론을 적용한 예시로서, PISA 2015 자료를 활용하여 학생이 인식한 과학의 도구적 동기 및 즐거움에 대해 국가별 측정동일성 검정 절차를 분석하고 국가별 잠재평균을 추정하였다. 마지막으로 본 연구의 향후 연구 및 의의에 대해 논의하였다.

In multi-group analysis, measurement invariance is a requirement for meaningful comparisons between groups. Multi-group confirmatory factor analysis (MG CFA) has been widely used for group comparisons. However, MG CFA is appropriate for comparisons with a small number of groups and is limited for a large number of groups, in particular, in cross-cultural comparative studies. To overcome the limitation of MG CFA, this study described alternative approaches: multilevel confirmatory factor analysis (ML CFA) and multilevel factor mixture modeling (ML FMM), which are effective for comparing more than 10 groups. In ML CFA, its advantages and disadvantages were described by introducing two models: random intercept models that estimate only intercepts as random effects and random intercept and loading models that estimate intercepts and factor loadings as random effects. Specifically, this study presented theoretical models for the two methods and procedures for testing measurement invariances. In addition, this study discussed advantages of ML CFA, relative to those of MG CFA, and several points that should be considered when applying ML CFA. And, as an example of applying ML CFA, this study conducted latent means analysis across countries for instrumental motivation of science and enjoyment perceived by students using the PISA 2015 data. Finally, implications of this study and future research directions were discussed.

Purpose: The objective of this study was to develop a predictive model for the sexual experiences of adolescents using the random forest method and to identify the "variable importance." Methods: The study utilized data from the 2019 to 2021 Korea Youth Risk Behavior Web-based Survey, which included 86,595 man and 80,504 woman participants. The number of independent variables stood at 44. SPSS was used to conduct Rao-Scott χ² tests and complex sample t-tests. Modeling was performed using the random forest algorithm in Python. Performance evaluation of each model included assessments of precision, recall, F1-score, receiver operating characteristics curve, and area under the curve calculations derived from the confusion matrix. Results: The prevalence of sexual experiences initially decreased during the COVID-19 pandemic, but later increased. "Variable importance" for predicting sexual experiences, ranked in the top six, included week and weekday sedentary time and internet usage time, followed by ease of cigarette purchase, age at first alcohol consumption, smoking initiation, breakfast consumption, and difficulty purchasing alcohol. Conclusion: Education and support programs for promoting adolescent sexual health, based on the top-ranking important variables, should be integrated with health behavior intervention programs addressing internet usage, smoking, and alcohol consumption. We recommend active utilization of the random forest analysis method to develop high-performance predictive models for effective disease prevention, treatment, and nursing care.

본 연구는 현재 유전평가에 사용하는 모델보다 많은 장점을 지니고 있는 임의회귀 검정일 모형(Random regression test-day model)을 이용할 때 마지막 세대의 불완전한 검정기록이 유전능력에 어떤 영향을 주는지 알아보고자 실시하였다.이용된 재료는 유우군능력검정사업을 통하여 수집된 2000년 1월부터 2005년 6월까지의 825,157개의 초산의 검정일 자료를 이용하였으며, 유전모수와 종모우의 육종가 추정은 REMLF90, BLUPF90을 이용하였다.

The data were collected in the dairy herd improvement program from January 2000 to July 2005. Test data included 825,157 records of first parity and animals with both parents known were included. This study aimed to describe the effect of incomplete lactation records of latest generation to the change in sire's breeding value using Random Regression model (RRM) in genetic evaluation. Estimation of genetic parameter and breeding value for sire used REMLF90 and BLUPF90 program. The phenotypic value on the number of test day records between group TD11, TD8, TD5, TD2 showed no large differences. For all the group heritability of test day milk yield range from 0.30 to 0.36. However TD2 group showed low heritability the least test day recode on the latest generation. The correlation of above 50% between test day and TD11(0.610), TD8(0.616), TD5(0.661) and TD2(0.682) with different records in latest generation. Sire's rank of breeding value varied widely depending on the records on the number of lactation from start to the latest generation. Study showed that change in breeding value ranked if daughter's test recode more so it should have at least 5 test day records. The use of RRM in dairy cattle genetic evaluation would be desirable if complete lactation records for latest generation daughters of young bulls when selection for proven bulls. Random Regression model (RRM) require at least 5 test-day lactation recode.

불안정한 기후와 함께 나타나는 국지적 집중호우로 인한 도시 침수는 끊임없이 발생하고 있으나, 강우량을 포함한 기상정보 현황 또는 예보정보를 활용하여 공간적인 도시홍수 예측정보를 제공할 수 있는 체계는 아직 마련되지 못한 상황이다. 공간적인 홍수정보는 하천의 제방, 도시 하수관거의 통수능, 저류지, 펌프시설과 같은 구조물적 대책에 어려움이 있을 시 발생할 수 있는 최악의 홍수상황을 미리 파악함으로써 피해를 최소화하는데 직접적인 영향을 미칠 수 있다. 이에 본 연구에서는 기상청에서 제공되는 강수량, 도시 유역에 대한 2차원 침수해석 결과, 그리고 기계학습 모형 중 하나인 랜덤포레스트 회귀모형을 활용하여 실시간으로 도시유역에 대한 침수지도를 예측할 수 있는 방법론을 제시하고자 한다. 연구유역은 내수침수가 빈번하게 발생하는 울산시 우정태화지구로 선정하였다. 지속시간 6시간의 총강우량 50 mm, 80 mm 그리고 110 mm 대한 랜덤포레스트 회귀분석 예측 침수면적과 검보정된 2차원 물리모형의 침수해석 결과 비교시 각각 63%, 80%, 그리고 67%의 적합도를 보여주어, 빠른 시간안에 발생하는 도시 침수에 대한 대응, 대피를 위한 기초자료로 활용될 수 있을 것으로 판단된다.

Urban flooding caused by localized heavy rainfall with unstable climate is constantly occurring, but a system that can predict spatial flood information with weather forecast has not been prepared yet. The worst flood situation in urban area can be occurred with difficulties of structural measures such as river levees, discharge capacity of urban sewage, storage basin of storm water, and pump facilities. However, identifying in advance the spatial flood information can have a decisive effect on minimizing flood damage. Therefore, this study presents a methodology that can predict the urban flood map in real-time by using rainfall data of the Korea Meteorological Administration (KMA), the results of two-dimensional flood analysis and random forest (RF) regression model. The Ujeong district in Ulsan metropolitan city, which the flood is frequently occurred, was selected for the study area. The RF regression model predicted the flood map corresponding to the 50 mm, 80 mm, and 110 mm rainfall events with 6-hours duration. And, the predicted results showed 63%, 80%, and 67% goodness of fit compared to the results of two-dimensional flood analysis model. It is judged that the suggested results of this study can be utilized as basic data for evacuation and response to urban flooding that occurs suddenly.

PURPOSES : The purpose of this study is to develop a crash prediction model at signalized intersections, which can capture the randomness and uncertainty of traffic accident forecasting in order to provide more precise results. METHODS : The authors propose a random parameter (RP) approach to overcome the limitation of the Count model that cannot consider the heterogeneity of the assigned locations or road sections. For the model's development, 55 intersections located in the Daejeon metropolitan area were selected as the scope of the study, and panel data such as the number of crashes, traffic volume, and intersection geometry at each intersection were collected for the analysis. RESULTS : Based on the results of the RP negative binomial crash prediction model developed in this study, it was found that the independent variables such as the log form of average annual traffic volume, presence or absence of left-turn lanes on major roads, presence or absence of right-turn lanes on minor roads, and the number of crosswalks were statistically significant random parameters, and this showed that the variables have a heterogeneous influence on individual intersections. CONCLUSIONS : It was found that the RP model had a better fit to the data than the fixed parameters (FP) model since the RP model reflects the heterogeneity of the individual observations and captures the inconsistent and biased effects.

이 연구는 확률적 모수를 고려한 음이항회귀분석을 이용하여 고속도로에서의 사고와 기하구조와의 관계를 파악하는데 목적이 있다. 고속도로에서의 기하구조는 모든 구간에 동일한 설계요소가 적용되기에는 현실적으로 불가능함에도 불구하고, 지금까지의 연구에서는 모형을 통해 도출되는 계수값이 구간에 설치된 기하구조의 특성에 관계없이 항상 고정된 값으로 추정되어왔다. 고정된 값을 이용한 일반적인 음이항모형은 시간적 변화 또는 각 대상구간이 가지고 있는 고유한 특성에 따른 변화를 통합하여 설명하지 못한다는 단점이 있으며, 이로 인해 추정된 계수의 표준오차가 과소추정되어 t-값이 부풀려지게 되며, 그 결과 모형의 설명력이 떨어지게 된다. 따라서, 이 연구에서는 워싱턴 주에 위치하고 있는 7개의 고속도로에서 발생한 9년동안의 사고자료 및 기하구조자료를 이용하여 구간별로 상이한 기하구조가 사고에 미치는 영향을 알아보고자 한다. 총 16개의 기하구조 관련 변수가 모형 도출에 이용되었으며, 기존의 음이항모형과의 비교를 통해 이 연구에서 제시하는 모형이 교통사고와 기하구조와의 관계파악에 더욱 적합함을 보이고자 한다. 그리고, 각 변수의 한계효용 및 탄력성 분석을 통해 이질성을 가지는 기하구조가 사고에 미치는 영향을 제시하고자 한다. 이는 향후 기하구조 관련 정책수립에 도움이 될 것으로 판단된다.

The objective of this study is finding the relationship between interstate highway accident frequencies and geometrics using Random Parameter Negative Binomial model. Even though it is impossible to take account of the same design criteria to the all segments or corridors on the road in reality, previous research estimated the fixed value of coefficients without considering each segment's characteristic. The drawback of the traditional negative binomial is not to explain the integrated variations in terms of time and the distinct characters specific segment has. This results in under-estimation of the standard error which inflates the t-value and finally, affects the modeling estimation. Therefore, this study tries to find the relationship of accident frequencies with the heterogeneous geometrics using 9-years and 7-interstate highway data in Washington State area. 16-types of geometrics are used to derive the model which is compared with the traditional negative binomial Model to understand which Model is more suitable. In addition, by calculating marginal effect and elasticity, heterogeneous variables' effect to the accidents are estimated. Hopefully, this study will help to estiblish the future policy of geometrics.

본 논문은 Even-Mansour 암호에 대해 안전성 약화 없이 키 사이즈를 줄이는 방법에 대해 다룬다. Even과 Mansour는 랜덤 순열 모델에서 랜덤 순열 P와 두 개의 키를 이용하여 평문 M을 암호화하는 기법($C=k_2\bigoplus P(M\bigoplus k_1)$)을 제안하였다. ASIACRYPT 2004에서 Gentry와 Ramzen은 4 라운드의 Feistel 구조를 이용하여 Even-Mansour 모델의 랜덤 순열을 랜덤 함수로 대치한 새로운 모델을 제안하고 안전성을 증명하였다. 본 논문에서는 Gentry-Ramzen 모델에 필요한 키 사이를 반으로 줄이는 방법을 살펴보고 제안한 방법에 대한 안전성을 랜덤 함수 모델에서 증명한다.

We describe the problem of reducing the key material in the Even-Mansour cipher without security degradation. Even and Mansour proposed a block cipher based on XORing secret key material just prior to and after applying random oracle permutation P such that $C=k_2\bigoplus P(M\bigoplus k_1)$. Recently, Gentry and Ramzan showed that this scheme in the random permutation oracle can be replaced by the four-round Feistel network construction in the random function oracle and also proved that their scheme is super-pseudorandom. In this paper we reduce the key size from 2n to n, which is the optimal key size of Even-Mansour cipher in the random function oracle model and also give almost the same level of security.

분류 문제에서 특정 범주의 빈도가 다른 범주에 비해 과도하게 높은 경우, 왜곡된 기계 학습을 유발할 수 있는 데이터 불균형(imbalanced data) 문제가 발생한다. 기업부도 예측 문제도 그 중 하나인데, 일반적으로 금융기관과 거래하는 기업들의 부도율은 대단히 낮아서, 부도 사례보다 정상 사례의 빈도가 월등히 높은 데이터 불균형 문제가 발생하고 있다. 이러한 데이터 불균형 문제를 해결하기 위해서는 적절한 표본추출 기법이 적용될 필요가 있으며, 지금껏 소수 범주 데이터를 복원 추출함으로써 다수 범주 데이터와 비율을 맞추어 데이터 불균형을 해결하는 오버 샘플링(oversampling) 기법이 주로 활용되어 왔다. 그러나 전통적인 오버 샘플링은 과적합화(overfitting)가 발생할 위험이 높아질 수 있는 단점이 있다. 이러한 배경에서 본 연구는 효과적인 기업부도 예측 모형 학습을 위한 표본추출 기법으로 2014년에 Menardi와 Torelli가 제안한 ROSE(random over sampling examples) 기법을 제안한다. ROSE 기법은 학습에 사용될 사례를 반복적으로 새롭게 합성하여 생성(synthetic generation)하는 기법으로, 과적합화 문제를 회피하면서도 분류 예측 정확도 개선에 도움을 줄 수 있다. 이에 본 연구에서는 ROSE 기법을 가장 성능이 우수한 이분류기로 알려진 SVM(support vector machine)과 결합하여 국내 한 대형 은행의 기업부도 예측에 적용해 보고, 다른 표본추출 기법들과의 비교연구를 수행하였다. 실험 결과, ROSE 기법이 다른 기법에 비해 통계적으로 유의한 수준으로 SVM의 예측정확도 개선에 기여할 수 있음을 확인하였다. 이러한 본 연구의 결과는 부도예측 외에 다른 사회과학 분야 예측문제의 데이터 불균형 문제 해결에도 ROSE가 우수한 대안이 될 수 있다는 사실을 시사한다.

If the frequency of a particular class is excessively higher than the frequency of other classes in the classification problem, data imbalance problems occur, which make machine learning distorted. Corporate bankruptcy prediction often suffers from data imbalance problems since the ratio of insolvent companies is generally very low, whereas the ratio of solvent companies is very high. To mitigate these problems, it is required to apply a proper sampling technique. Until now, oversampling techniques which adjust the class distribution of a data set by sampling minor class with replacement have popularly been used. However, they are a risk of overfitting. Under this background, this study proposes ROSE(Random Over Sampling Examples) technique which is proposed by Menardi and Torelli in 2014 for the effective corporate bankruptcy prediction. The ROSE technique creates new learning samples by synthesizing the samples for learning, so it leads to better prediction accuracy of the classifiers while avoiding the risk of overfitting. Specifically, our study proposes to combine the ROSE method with SVM(support vector machine), which is known as the best binary classifier. We applied the proposed method to a real-world bankruptcy prediction case of a Korean major bank, and compared its performance with other sampling techniques. Experimental results showed that ROSE contributed to the improvement of the prediction accuracy of SVM in bankruptcy prediction compared to other techniques, with statistical significance. These results shed a light on the fact that ROSE can be a good alternative for resolving data imbalance problems of the prediction problems in social science area other than bankruptcy prediction.