• Genomics & Informatics
• /
• 제17권4호
• /
• pp.41.1-41.12
• /
• 2019

Survival analysis mainly deals with the time to event, including death, onset of disease, and bankruptcy. The common characteristic of survival analysis is that it contains "censored" data, in which the time to event cannot be completely observed, but instead represents the lower bound of the time to event. Only the occurrence of either time to event or censoring time is observed. Many traditional statistical methods have been effectively used for analyzing survival data with censored observations. However, with the development of high-throughput technologies for producing "omics" data, more advanced statistical methods, such as regularization, should be required to construct the predictive survival model with high-dimensional genomic data. Furthermore, machine learning approaches have been adapted for survival analysis, to fit nonlinear and complex interaction effects between predictors, and achieve more accurate prediction of individual survival probability. Presently, since most clinicians and medical researchers can easily assess statistical programs for analyzing survival data, a review article is helpful for understanding statistical methods used in survival analysis. We review traditional survival methods and regularization methods, with various penalty functions, for the analysis of high-dimensional genomics, and describe machine learning techniques that have been adapted to survival analysis.

• 한국농공학회논문집
• /
• 제59권5호
• /
• pp.101-108
• /
• 2017

Recently, the occurrence of landslides has been increasing over the years due to the extreme weather event. Developments of landslides monitoring technology that reduce damage caused by landslide are urgently needed. Therefore, in this study, a strain ratio sensor was developed to predict the ground behavior during the slope failure, and the change in surface ground displacement was observed as slope failed on the field model experiment. As a result, in the slope failure, the ground displacement process increases the risk of collapse as the inverse displacement approaches zero. It is closely related to the prediction of precursor. In all cases, increase in displacement and reverse speed of inverse displacement with time was observed during the slope failure, and it is very important event for monitoring collapse phenomenon of risky slopes. In the future, it can be used as disaster prevention technology to contribute in reduction of landslide damage and activation of measurement industry.

• 한국수자원학회논문집
• /
• 제54권9호
• /
• pp.667-680
• /
• 2021

이 연구의 목적은 서리 발생일과 무상일 기간의 특성을 분석하고 로지스틱 회귀, 랜덤 포레스트, Long-short Term Memory (LSTM) 기법을 활용하여 서리발생 예측모델을 개발하고 평가하는데 있다. 수원, 청주, 광주 지점에서 봄철과 가을철 서리발생 예측모델 개발을 위한 기상변수들을 수집하였으며, 수집기간은 1973년부터 2019년까지이다. 프리시전(precision), 리콜(Recall), f-1 스코어와, AUC 및 Reliability Diagram과 같은 그래피컬 평가기법을 이용해 서리발생 예측모델을 평가하였다. 봄철과 가을철 모두 서리발생일이 줄어드는 경향성(유의수준: 0.01)을 보였다. 0.9 이상의 높은 AUC 값에도 불구하고, 신뢰도는 일정한 값을 보여주지는 않았다. 서리발생일 측뿐만 아니라, 초상일과 종상일을 정확히 예측할 수 있도록 모형 개선이 필요해 보이며, 다른 지역의 더 많은 지점에서 동일한 기법을 적용해 보는 연구가 필요해 보인다.

• 한국컴퓨터정보학회논문지
• /
• 제29권2호
• /
• pp.51-59
• /
• 2024

투자자는 수익의 극대화를 위해 언론사의 기사를 포함한 다양한 정보를 활용하여 투자 전략을 수립한다. 이에 국내 언론사에서도 신뢰도 있는 투자정보를 제공하기 위해, 애널리스트의 종목분석 보고서에 기초한 종목 추천기사를 게재하고 있다. 본 연구에서는 종목 추천기사 게재를 하나의 사건(event)으로 간주하고, XGBoost와 LightGBM 모델을 활용하여 기사 게재 10일 이후 가격의 상승 또는 하락을 예측하는 분류 모델을 제시한다. 또한, 전체 추천종목을 유가증권시장과 코스닥 시장 및 기업규모(대형/소형)에 따라 4가지로 분류하고, 하위 그룹에 따라 모델의 예측 정확도에 차이가 있는지 파악하고자 한다. 학습 결과 전체 모델의 분류 정확도는 XGBoost 75%, LightGBM 71%로 나타났고, 예측 정확도는 유가증권 시장 예측력이 코스닥시장 주식 대비 높게 나타났으며, 대형주의 예측력이 소형주 보다 높게 나타났다. 마지막으로, SHAP(Shapley Additive exPlanations) 분석을 통해 개별 모델의 예측에 중요한 변수를 살펴보고 모델의 해석력을 제고하였다.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1997년도 추계학술대회 논문집 학회본부
• /
• pp.245-249
• /
• 1997

A feasibility study on the characterization of incipient insulator failure for distribution fault prediction is presented. In this study, real distribution data was collected and analyzed to isolate incipient failure signatures or parameters which were expected to show distinct behaviors before and after failure incident. Several signal analysis methods were applied to isolate the parameters and a new strategy of analysis, the event-date concept, was also applied to find a relationship between non-harmonic and high frequency signal activities and imminent insulator failures.

• 한국경영과학회지
• /
• 제28권2호
• /
• pp.105-127
• /
• 2003

In this paper, four different data mining techniques, two neural networks and two statistical modeling techniques, are compared in terms of prediction accuracy in the context of bankruptcy prediction. In business setting, how to accurately detect the condition of a firm has been an important event in the literature. In neural networks, Backpropagation (BP) network and the Kohonen self-organizing feature map, are selected and compared each other while in statistical modeling techniques, discriminant analysis and logistic regression are also performed to provide performance benchmarks for the neural network experiment. The findings suggest that the BP network is a better choice among the data mining tools compared. This paper also identified some distinctive characteristics of Kohonen self-organizing feature map.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1998년도 하계학술대회 논문집 C
• /
• pp.1213-1216
• /
• 1998

This paper presents the result of a feasibility study for the prediction method of fault symptoms on 22.9kV distribution line. In this paper, real distribution data was collected and analyzed to isolate failure signatures or parameters which were distinct behaviors before and after failure incident. A new strategy of analysis-based (event-date concept) prediction algorithm for the distribution insulators and a developed model system were also discussed.

• 융합보안논문지
• /
• 제16권2호
• /
• pp.63-70
• /
• 2016

최근 인터넷 등 정보통신 기술의 발달로 인해 언제 어디서나 인터넷을 이용할 수 있는 환경이 구축 되었으며, 이로 인한 사이버위협은 다양한 경로를 통해 시도되고 있다. 본 연구에서는 사이버위협 중 지속적으로 증가 추세인 DDoS 예측 모델링하기 위해 이벤트 데이터를 근거로 하여 통계적 기법을 통해 DDoS 위험지수 예측식을 도출하였고, 도출된 위험지수를 정량화하였다. 제시된 위험지수를 활용하여 DDoS 위협에 대해 사전 대응정책을 세움으로써 피해를 최소화시킬 수 있는 객관적이고 효율적인 예측 모델이 될 것으로 기대한다.

• Ocean and Polar Research
• /
• 제31권4호
• /
• pp.369-377
• /
• 2009

Performance of the Korea Ocean Research and Development Institute (KORDI) operational storm surge prediction system for the Korean coast is presented here. Results for storm surge hindcasts and forecasts calculations were analyzed. The KORDI storm surge system consists of two important components. The first component is atmospheric models, based on US Army Corps of Engineers (CE) wind model and the Weather Research and Forecasting (WRF) model, and the second components is the KORDI-storm surge model (KORDI-S). The atmospheric inputs are calculated by the CE wind model for typhoon period and by the WRF model for non-typhoon period. The KORDI-S calculates the storm surges using the atmospheric inputs and has 3-step nesting grids with the smallest horizontal resolution of ${\sim}$300 m. The system runs twice daily for a 72-hour storm surge prediction. It successfully reproduced storm surge signals around the Korean Peninsula for a selection of four major typhoons, which recorded the maximum storm surge heights ranging from 104 to 212 cm. The operational capability of this system was tested for forecasts of Typhoon Nari in 2007 and a low-pressure event on August 27, 2009. This system responded correctly to the given typhoon information for Typhoon Nari. In particular, for the low-pressure event the system warned of storm surge occurrence approximately 68 hours ahead.

• Wind and Structures
• /
• 제30권4호
• /
• pp.393-403
• /
• 2020

One of the growing concerns of the wind energy production is wind ramp events. To improve the wind ramp event forecasts, the nonlinear Kalman filter bias correction method was applied to 24-h wind speed forecasts issued from the WRF model at 70-m height in Zhangbei wind farm, Hebei Province, China for a two-year period. The Kalman filter shows the remarkable ability of improving forecast skill for real-time wind speed forecasts by decreasing RMSE by 32% from 3.26 m s_-1 to 2.21 m s_-1, reducing BIAS almost to zero, and improving correlation from 0.58 to 0.82. The bias correction improves the forecast skill especially in wind speed intervals sensitive to wind power prediction. The fact shows that the Kalman filter is especially suitable for wind power prediction. Moreover, the bias correction method performs well under abrupt weather transition. As to the overall performance for improving the forecast skill of ramp events, the Kalman filter shows noticeable improvements based on POD and TSS. The bias correction increases the POD score of up-ramps from 0.27 to 0.39 and from 0.26 to 0.38 for down-ramps. After bias correction, the TSS score is significantly promoted from 0.12 to 0.26 for up-ramps and from 0.13 to 0.25 for down-ramps.