순차적 경기수행력 분석을 위한 순환신경망의 적용 - 실업검도 대회를 중심으로

박지훈(Jihoon PARK) 한국체육측정평가학회 2020 한국체육측정평가학회지 Vol.22 No.4

스포츠 현장에서 경기의 승ㆍ패를 예측하는 것은 개인과 팀의 경기력 향상이라는 상호작용 효과를 준다. 데이터를 처리하는 방법의 새로운 기술개발로 단순한 정보처리에서 다차원의 복잡한 신경망까지 일반 개인이 구축할 수 있는 환경이 되었으며, 적용 가능성을 시험하기 위한 검증 무대로서 가장 적합한 스포츠 현장에서 다양한 시도가 계속되고 있다. 검도는 단체스포츠가 아닌 개인 간 무도 종목이다. 동일한 조건에서 정해진 부위를 타격하여 득점하는 경기이며, 타 종목에 비해 경기상황의 경우의 수가 적다. 이 같은 이점을 활용하여 본 연구에서는 검도경기의 승ㆍ패 예측을 기계학습을 통해 추정하고자 하였으며, 구체적으로 검도선수들의 검도 경기수행력을 순차적으로 부호화 시키고 득점상황을 예측하는데 목적이 있다. 사용된 데이터는 2019년 실업검도대회 3개를 대상으로 학습데이터셋과 교차타당도검증 데이터셋으로 분류하였다. 경기분석 프로그램인 Longomatch 1.6을 사용하여 경기영상을 통해 자료를 부호화 하였고, 기계학습을 위해 Tensor 형태로 변환하는 전처리 과정 및 순환신경망(LSTM) 모델 개발을 통계프로그램 R 3.6.1 프로그램을 활용하였다. 연구결과 개발된 다양한 신경망 중 64-128-64-32-3 레이어 신경망이 학습정확도 88.5%의 예측률을 보였다. 그리고 순환신경망(LSTM)을 활용하여 경기 내용을 학습시키고 실제 경기에서 적용하였을 경우 득점 여부를 79% 예측할 수 있었다. 결론적으로 본 연구에서 개발된 예측모델은 선수나 지도자가 경기내용의 평가 및 훈련에서 경기력 향상을 위해 다양하게 활용될 수 있을 것이다. 추후 무도종목을 비롯한 다양한 대인종목에서도 순환신경망의 적용 가능성을 기대한다. Modern society is rapidly developing as confirming the analysis and applicability of big data. With the development of new technologies in the data processing method, it has become an environment that individual can build from simple information processing to multidimensional complex neural networks, and various attempts continue at the most suitable sports fields as a proving ground to test its applicability. Kendo is an individual martial arts sport not a group sport, which player can score points by hitting the appointed areas under the same conditions, and there are fewer number of cases in the game compared to other sports. To take this advantages, it was about to develop the scoring prediction algorithm which affect the outcome of the game and the used data was classified into learning dataset and cross validity validation dataset and developed LSTM algorithm for the three business Kendo Championships. The data was coded through competition video and the preprocessing process that converting to Tensor was conducted for machine learning. 64-128-64-32-3 layer neural network among the various developed neural network showed 88.8% learning accuracy, 88.5% of cross-validation accuracy, and 78.5% of prediction rate. By using LSTM, it could predict 79% of the scoring when learning the contents of Kendo and applying in the real game, and the players and it will be used that players and instructors can use it for the assessment of the game and training in the various ways to improve the performance. If it is added more data later and supplemented the algorithm, it can be used as grounded basic data generation and supplementary data of VAR function regarding the Kendo matches studied academically.

건설 경기종합지수를 활용한 공종별 건설경기 예측

박철한 ( Cheol Han Park ) 한국건설경제산업학회 2021 건설경제산업연구 Vol.8 No.2

지난 2020년부터 국내 경제는 코로나19로 인하여 불확실성이 높아진 상황으로 공종별 건설경기 상황을 정확히 판단하고 예측할 필요가 있어 본 연구를 진행하였다. 과거 연구를 참고 건설시장을 건설 생산시장, 건설 노동시장, 건설 금융시장, 건설 자재시장, 그리고 건설 소비시장인 부동산시장으로 구분하여 관련 시계열 통계를 취합했다. 경제적 중요성(Economic Significance), 통계적 적합성(Statistical Adequacy), 경기 속보성(Currency), 경기 대응성(Conformity)을 변수 선정의 기준으로 평가하였으며, 공종별 건설투자에 대응한 수준변수, 로그차분변수, 순환변동치 그리고 전년 동기비에 대한 시차상관변수 분석을 수행해 종합지수 선정에 사용되는 지표를 선정하였다. 공종별 경기종합지수는 일반적인 통계청(NBER)이 사용하는 방법과 동일한 방법을 통해 지수화해 분석을 수행했다. 2000년 1월∼2021년 6월까지 데이터를 토대로 공종별 경기동행지수와 공종별 경기선행지수를 토대로 공종별 전망모형을 구축한 이후 이를 해석할 보조 지표를 HP 필터와 마코프 국면전환 확률을 사용했다. 이후 1년치 전망(2021년 7월∼2022년 6월)을 수행하였다. 선정된 공종별 동행지수와 선행지수를 ARIMA 모형을 사용해 2021년 7월부터 2022년 6월까지 1년치를 전망하였으며 이를 통하여 최근 공종별 건설경기를 판단하고 예측치가 시사하는 바를 해석하였다. 특히, 선행지수의 순환 변동치를 마코프 국면 전환 모형을 이용해 침체국면(low regime) 확률을(Recession Probability) 도출해 향후 공종별 건설경기의 침체 여부를 판별하였다. 본 연구를 통한 1년 전망치가 제시하는 시사점은 건설경기 양호하지만 2022년 2/4분기 선행지수의 상승세가 둔화될 가능성을 제시하고 있는데 이는 예상했던 것보다 주거용 건축경기 침체 시점이 빨라질 수 있음을 시사하고 있으며, 비주택 건축과 토목 건설경기의 호조세가 2022년 상반기까지 지속될 것임을 시사하고 있다. Since 2020, in South Korea, the domestic economy is in a situation of increased uncertainty due to the continued re-proliferation of COVID-19. In such a situation, predicting the future construction market is an important and meaningful study. In this study, through the NBER method I made 4 composite leading indicator indexs and 4 coincident composite indexes, calssified construction type such as residential buildings, non-residential buildings, civil engineering and total construction, in order to understand domestic construction market by type. I forcasted each coincident composite indexes from July 2021 to June 2022 using ARIMA model with leading indicator data and analysed their fluctation through HP(Hodrick - Prescott) filer and calculated the recession probability through basic Markov swiching model. The implications of the one-year forecast through this study are as follows. First, although the construction market is healthy in 2021, it suggests the possibility of a transition period in the first half of next year, especially in the second quarter. Based on the data through June 2021, it suggests that the transition point of the residential construction market may proceed earlier than expected. Second, the transition point of the non-residential construction market is judged to be delayed compared to the residential construction market. It is expected that the non-residential construction market will adjust after the residential construction market is adjusted. Overall, it is analyzed that the increase in the non-residential construction market is likely to continue until the first half of 2022. Third, the recovery of the civil engineering construction industry is expected to be slower than that of the construction industry, and it is expected to be achieved in a gradual fashion. This study can be said to be an early study in examining and forecasting changes in the construction market by construction type. In order to accurately monitor economic changes, continuous follow-up studies are needed. Above all, it is necessary to study how rapidly changing financial and macro-environmental changes will affect the construction industry by construction type. In addition, research is needed to continuously update the forecast model prepared for each type of construction and improve it into an optimal model with high predictive power for each period.

기온효과를 고려한 건설업생산지수 예측모델 개발

김석종,김현우,진경호,장한익,Kim, Seok-Jong,Kim, Hyun-Woo,Chin, Kyung-Ho,Jang, Han-Ik 한국건설관리학회 2013 한국건설관리학회 논문집 Vol.14 No.5

1990년대 이후 국가경제에서 미치는 영향이 감소 추세에 들어선 건설업은 호황과 불황을 넘나들고 있다. 건설업의 경기변동이 심할수록 경기예측은 어려워지며, 불확실한 예측의 피해는 기업과 건설 종사자들이 직접적으로 받게 되므로 건설경기를 예측하는 것은 매우 어려우면서 중요한 일이다. 본 연구에서는 건설경기를 나타내는 지표 중 하나인 건설업생산지수를 GDP와 기온효과를 이용하여 실질소득과 야외활동이 많은 건설업의 특성에 따라 기온효과를 반영한 공급측면에서의 단기 건설 경기예측 모형을 제시하였다. 분석결과, 건설경기는 뚜렷한 기온효과가 있으며 GDP에도 큰 영향을 받는 것으로 나타났다. 이와 같은 과정을 통해 입증된 건설경기 예측모델을 기반으로 GDP예상증가율 3.5%와 2.4%일 때, 두 가지 시나리오로 2013년도 건설업생산지수를 예측하였다. 본 연구결과는 건설업의 경기를 판단하는 지표 중 하나로 활용 가능할 것이며, 향후 기후변화가 건설업에 미치는 영향에 대한 연구의 초석이 될 것이다. After 1990s, the influence of construction industry has been decreased on national economy and construction business condition has been changed on economic recession and boom repeatedly. Larger fluctuation of business condition makes a forecast of it to be more difficult. Uncertainty in business prediction results in damages on construction companies and stakeholders. Therefore, study on forecasting a construction business is very important. This study suggests the Construction Industry Production Index(CIPI) to predict a construction business in consider of temperature effects. The results show that construction business is much influenced by temperature effects certainly and GDP. With the CBFM, this study examines CIPI for 2013 with two scenarios: 1)with GDP growth rate of 3.5% 2)with GDP growth rate of 2.4%. Thus, CIPI would be used as the economic state index to display the construction business conditions. Also, CIPI will be utilized as basic methodology in the impact of climate change in the construction industry.

생명보험산업 보험료 성장률 예측계량모형 비교

전성주 ( Sungju Chun ),조영현 ( Younghyun Cho ) 한국금융연구원 2015 금융연구 Vol.29 No.3

In this article we evaluate the performance of forecasting models to predict the Korean life insurance premium growth rates. Comparisons are made for the Vector Auroregressive (VAR) predictive model, the multivariate leading indicator model, and the diffusion index model proposed by Stock and Watson (2002) against the Univariate Autoregressive (AR) predictive model as a benchmark. We compare each model’s predictability for the total premium incomes and the initial premium incomes of three types in the individual insurance; protection, endowment and annuity. The complete quarterly data spans from the second quarter of 1986 to the first quarter of 2014. The lag selection for AR and VAR forecasting models depends on the Bayesian Information Criteria (BIC) with the maximum number of lags set to 4. For the multivariate leading indicator model, we use 4 leading indicators of GDP growth rates, inflation, education, and age. In order to avoid data mining concerns, we select the variables that have been found to be the determinants of life insurance demands by previous studies. The diffusion index model is an approximate dynamic factor model that relates the future life insurance premium growth rates to a number of factors estimated by principal components using a large number of macroeconomic variables. The set of macroeconomic variables consists of 57 variables representing 6 main categories of macroeconomic time series: demand for final output; balance of payments and international trade; price indexes; money, interest rates and financial markets; labor, production and population; and world variables. We also include in the data set the variables related to the life insurance industry such as the industry’s total asset returns, claims paid and etc. In predicting total premium income growth rates, the AR predictive model produces the smallest mean squared predictive errors (MSPEs). For the premium incomes of protective insurance all the forecasting models have the MSPEs less than 3%. But they become unreliable in predicting the premium incomes of annuity with producing the MSPEs more than 10%. When we test the null hypothesis of no difference in MSPEs, it is rejected at 5% significance level for the VAR and the diffusion index predictive models when we forecast the total premium income growth rates of protective insurance. In predicting initial premium income growth rates, we find that there are no statistically significant differences in the MSPEs of each model. In addition, all the models have MSPEs more than 10% so that we may not be able to depend on any model to forecast in practice. We conclude that it may not be beneficial to take advantage of the information contained in macroeconomic variables for predicting life insurance premium growth rates. It may be due to the fact that most of the insurance contracts in Korea charge monthly premiums, which induces heavy autocorrelations among quarterly insurance premium data and makes an AR forecast very effective. Rho and Shin (1998) also found that macroeconomic variables are not likely to influence on insurance demands as is largely determined by insurer’s push-marketing. Policy holders cannot surrender their insurance contracts without heavy penalties, which makes them not so much responsive to macroeconomic environments. Lastly, life insurance demands are very sensitive to changes in insurance regulation and taxation, which could not be controlled for by our estimation procedure due to the lack of time series observations.

금리스프레드의 경기 예측력 비교분석

김민국,이한식 통계청 2019 통계연구 Vol.24 No.1

The usefulness of interest rate spreads for forecasting future economic activity is now well known in the macroeconomics literature. In fact, the term spread has been included as an indicator among the composite leading index of Korea since 2006. In this paper, we explore the predictive content of alternative spreads based on the probit analysis. First, we examine the business-cycle predictability of two versions of term spreads, where five-year and three-year government bond rates are used for long-term interest rates, while the call rate plays the role of short-term rate. We also compare the credit spread, derived as the difference between the return on corporate bold of AA- class and the government bond rate. The basic finding is that the term spreads have more predictive content for business-cycle fluctuations than the credit spread. Also presented is evidence that the term spread using five-year government bond rate leads to a better prediction for future recessions than that using three-year government bond rate. Such results are expected to contribute to the background for the recent revision of the composite economic indices in 2016. 금리스프레드는 경기변동의 예측에 중요한 정보변수로 활용되고 있다. 우리나라는 2006년 경기종합지수 개편 이후 장단기 금리스프레드를 경기선행지수 구성 항목으로 사용하고 있으며, 2016 년에 장기금리 자료를 3년 만기에서 5년 만기 국고채로 변경했다. 본 연구에서는 다양한 금리스프레드의 경기변동 예측력에 대한 비교?분석을 시도했다. 이를 위해 3년 및 5년 만기 국고채수익률과 콜금리의 차이로 정의된 두 기간스프레드와 3년 만기 회사채 수익률과 국고채 수익률간의 신용스프레드를 구성했다. 선형회귀모형 추정 결과에 의하면, 스프레드에 따라 선행성의 차이는 있지만 미래의 경기변동을 잘 예측하는 것으로 분석됐다. 프로빗 모형을 적용한 분석 결과, 각 스프레드가 미래 경기전환점을 어느 정도 잘 예측하는 것으로 나타났다. 특히 5년 만기 수익률을 사용한 스프레드가 3년 만기 수익률을 사용한 스프레드에 비해 예측력이 우수한 것으로 분석됐다. 이러한 결과는 장기금리 기초자료를 조정한 통계청의 최근 경기종합지수 개편 배경과일치하는 것으로 판단된다. 선행지표의 체계적 구성이 미래 경기예측에 중요한 의미를 갖는다는점을 고려할 때, 금리스프레드의 효용성을 재조명한 본 연구의 분석결과는 향후 경기종합지수운용에도 유용한 정보를 제공할 것으로 기대된다.

A Study on the Development of a Hotel Economic Trend Predictive Model Using Machine Learning Models

이규태,홍경옥 한국비교정부학회 2024 한국비교정부학보 Vol.28 No.1

(Purpose) Based on a total of 21 hotel economic trend data collected through the Korea Statistical Information Service (KOSIS) and the Korea Hotel Industry Association, this study uses a machine learning model to determine the prediction model estimation and the importance of variables to propose a differentiation strategy for hotels to preemptively respond to rapidly changing economic fluctuations. (Design/methodology/approach) Based on previous research, a machine learning model was used to grasp the prediction accuracy and importance of variables in the hotel economic trend model. Data collection was based on a total of 12 years of data from 2011 to 2022, and the data cycle was set on a monthly and quarterly basis. On the other hand, the orange data mining 3.32 program, a machine learning program, was used to grasp the prediction accuracy of hotel business trends. (Findings) The prediction accuracy was the highest in the linear form at 93.3%, followed by the neural network at 89.7%, linear at 83.3%, and support vector machine at 75.6%. In other words, it is judged that it is necessary to actively use adaboost and neural network models to predict hotel business trends and future time series models. (Research implications or Originality) In order to predict hotel economic trends, in addition to existing macroeconomic indicators, room profit rate indicators, which directly affect hotel revenue, should be considered. Implications of this study provide academic and managerial suggestions and limitations of the study and directions for future research are suggested.

경기변동에 따른 비상장 중소기업 신용위험을 설명하는 회계변수와 부도예측에 관한 실증연구

김성규 ( Sung Kyoo Kim ),이화득 ( Hwa Deuk Yi ) 한국회계학회 2012 회계저널 Vol.21 No.6

예측 모형 통합을 통한 호텔산업 경기 동향 예측 연구

김명준 (사)한국관광레저학회 2018 관광레저연구 Vol.30 No.3

Hotel industry demand has been rapidly changing by not only each countries economic condition but also relationship between the countries. The prediction model should consider various phenomena for the proper and accurate forecasting. Most of hotel industry forecasting indicators are produced by prediction models which are built by consideration of explanatory variable significances and data manipulations. Those models can be one of candidates for the prediction but have limitations of reflecting the specific changes especially diplomatic issues influencing the number of incoming tourists of certain country. This study suggests prediction model integration methodology which is more applicable hotel industry trends forecasting based on individual models for the each country. The suggested method could reflect the specific issue just related with certain country and the integration idea also could be expanded in various fields as an alternatives.

호텔산업 경기 동향 예측 모델에 관한 연구

한창엽(Han, Chang-Youp),김명준(Kim, Myung Joon) 한국관광레저학회 2016 관광레저연구 Vol.28 No.9

벌집순환모형을 아파트시장 분석 및 이용한 지역별 예측에 관한 연구

마승렬 ( Seungryul Ma ) 주택도시보증공사 2016 주택도시금융연구 Vol.1 No.1

Conducting an accurate analysis and providing a reasonable forecast of regional housing cycles are essential to stabilize the housing market because the uncertainties in the housing market can have an enormous impact on dwellings and national economy. This paper analyzes regional housing cycles using HCM (Honeycomb Cycle Model) and forecasts the future housing cycles. While HCM has long been used in the analysis of the cycles of Korean housing market, some limitations of the model have been pointed out by other studies. We re-evaluated the suitability of HCM as an analysis tool for regional housing cycles using the data of housing sales volumes and real house prices in Korea. As a result, we confirmed that HCM is a useful tool for analysis and forecast of regional housing cycles in Korea.