http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
池元哲 弘益大學校 科學技術硏究所 2005 科學技術硏究論文集 Vol.16 No.-
Fraud detection is a special case of classification problem in that the target class sample size is extremely small compared to its opposite class. Simple oversampling technique dose hardly contribute to improving the classification accuracy, due to typical overfitting phenomenon. Two approaches can be considered to effectively address this knid of probelms. First, lots of derived variables are generated and tested to find the set of input variables that empowers the classifier to solve this kind of classification problem. The other is to adopt a certain kind of classifier that shows excellent performance for this problem without special eattention to input variables. In this paper, Support Vector Machine(SVM) is adopted in an effort to find new powerful classifier for fraud detection problem. Credit card transaction data is used to test the classification performance of SVM that showed the promising results. SVM needs to be, however, rigoriously tested on the large volume of data because of its training time and the current limitation to the size of training set.
施設再配置를 허용한 動的 Location Problem에 關한 硏究
池元哲 弘益大學校 1983 弘大論叢 Vol.15 No.2
In the area of dynamic facility location problems, one desires to select the time-staged establishment of facilities at different locations, so as to minimize the total discounted costs for meeting demands specified over time at various customer locations. A particular dynamic facility location problem is formulated as a combinatorial optimization problem. The formulation permits the relocation of facilities which are assumed to be capacity-unlimited A branch and bound procedure incorporating a Lagrangean relaxation method is presented to solve this problem, which is reformulated as a shortest path problem to find the lower bound of branch and bound procedure.
池元哲,趙相喆 弘益大學校 科學技術硏究所 2001 科學技術硏究論文集 Vol.12 No.-
Association Rule Mining,(Agrawal et al., 1993) is an exploration method which searches associative relations between item sets in large database. Association Rule Mining, in general, searches association between qualitative items. To adapt Association Rule Mining to quantitative attributes, Quantitative Association Rule Mining is introduced in 1996. Sequential Pattern Mining and N-Dimensional Inter Transaction Mining are applications of Association Rule for Time series analysis by extending time dimensions to the Association Rule. Analyzing complex time series, such as stock price movement, using quantitative model is limited, and Technical Analysis like a chart analysis is an alternative approach for them. Technical Analysis recognizes patterns, and analyzes impacts effected by the pattern. In this work, we defined pattern on the time series syntactically and analyzed the patterns and recognized the patterns which is frequently emerges in the time series, and, explored the other patterns as an impact of discovered patterns in former step. For this purpose, we used N-dimensional Inter Transaction Association Rule and syntactically described pattern(of both quantitative, qualitative attribute). So it enables pattern recognition, analyzing impacts and forecasting of complex time series such as stock prive movement
프로세스 플래닝의 자동화를 위한 각주형 파트의 특징형상 인식: 확장된 AAG 접근 방법
池元哲 弘益大學校 科學技術硏究所 1995 科學技術硏究論文集 Vol.6 No.-
This pape describes an approach to recognizing composite features of prismatic parts. AAG(Attribute Adjacency Graph) is adopted as the basis of describing basic features, but Extended AAG (EAAG) is suggested to enhance the expressive power of AAG by adding face type. angles between faces and normal vectors. To simplify the recognition procedure, features classification tree is built using the graph types of EAAG and the number of EAD`s. Algorithms to find open faces and dimensions of features is exemplified and used in decomposing composite features. The processing sequence of recognized features is automatically determined during the decomposition process of composite features.
池元哲 弘益大學校 1988 弘大論叢 Vol.20 No.2
This paper discusses the design of expert system for financial statement analysis. The concept of expert support system is introduced and adopted as a framework for system development. Combined Knowledge representation scheme-frame and rule, and model-based quantitative reasoning capability based on simple financial arithmetics will greatly enhance the system flexibility and ensure wide applicability to real business problem.
池元哲 弘益大學校 1989 弘大論叢 Vol.21 No.2
In this paper, the applicability of connectionist network is tested against CP rating. A three-layered Connectionist network is designed, and of which performance in learning agency judgement of CP rating is assessed, and compared with prediction model derived from multiple discriminant analysis. The results show that connectionist network outperforms MDA, and can be use to develop a more accurate prediction model for CP rating.