• Journal of the Korean Institute of Intelligent Systems
• /
• v.25 no.2
• /
• pp.191-196
• /
• 2015

There have been many studies on statistical forecasting on firm's performance and stock price by applying various financial indicators such as debt ratio and sales growth rate. Selecting predictors for constructing a prediction model among the various financial indicators is very important for precise prediction. Most of the previous studies applied variable selection algorithms for selecting predictors. However, the variable selection algorithm is considered to be at risk of eliminating certain amount of information from the indicators that were excluded from model construction. Therefore, we propose a firm's performance prediction model which principal component analysis is applied instead of the variable selection algorithm, in order to reduce dimensionality of input variables of the prediction model. In this study, we constructed the proposed prediction model by using financial data of American IT companies to empirically analyze prediction performance of the model.

• Journal of KIISE:Computing Practices and Letters
• /
• v.6 no.6
• /
• pp.555-568
• /
• 2000

The World Wide Web has become the largest virtual system that is almost universal in scope. In recent research, it has become effective to utilize idle hosts existing in the World Wide Web for running applications that require a substantial amount of computation. This novel computing paradigm has been referred to as the advent of global computing. In this paper, we implement and propose a mobile object-based global computing framework called Tiger, whose primary goal is to present novel object-oriented programming libraries that support distribution, dispatching, migration of objects and concurrency among computational activities. The programming libraries provide programmers with access, location and migration transparency for distributed and mobile objects. Tiger's second goal is to provide a system supporting requisites for a global computing environment - scalability, resource and location management. The Tiger system and the programming libraries provided allow a programmer to easily develop an objectoriented parallel and distributed application using globally extended computing resources. We also present the improvement in performance gained by conducting the experiment with highly intensive computations such as parallel fractal image processing and genetic-neuro-fuzzy algorithms.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.19 no.1
• /
• pp.35-40
• /
• 2015

The paper proposes a meta-learning approach for building multi-classifier systems in a GA-based inductive learning environment. In our meta-learning approach, a classifier consists of a general classifier and a meta-classifier. We obtain a meta-classifier from classification results of its general classifier by applying a learning algorithm to them. The role of the meta-classifier is to evaluate the classification result of its general classifier and decide whether to participate into a final decision-making process or not. The classification system draws a decision by combining classification results that are evaluated as correct ones by meta-classifiers. We present empirical results that evaluate the effect of our meta-learning approach on the performance of multi-classifier systems.

• Journal of Navigation and Port Research
• /
• v.27 no.4
• /
• pp.397-402
• /
• 2003

Generally, container arrangement has been carried out based on the Classification guidelines. However, guidelines provide only container securing forces for the given container arrangement and Classification requirements of the forces. In order to design container arrangement additional information is needed such as container securing forces and arrangement that accounts for lashing bridges, vertical lashing, vertical center of gravity (VCG), and maximum stack weight. Trial and error method using the existing guidelines requires excessive amount of calculation time and cannot provide accurate results of the calculations. In order to fulfill this need, a new container securing system has been established based on the equilibrium conditions that include lashing bridges and vertical lashing. An optimization algorithm has been developed for the new system since current optimization methods such as genetic algorithms and evolution strategies are unsuitable for the container securing problems, which involve equality constraint. Design variables are container weights of tier and objective function is either total container weight or VCG of a stack. The newly developed system provides optimum arrangement of containers for both maximum stack weight and maximum VCG. It also greatly reduces time for designing container arrangement.

• Journal of Intelligence and Information Systems
• /
• v.11 no.2
• /
• pp.175-195
• /
• 2005

사례기반추론(case-based reasoning)은 사례간 유사도를 평가하여 유사한 이웃사례를 찾아내고, 이웃사례의 결과를 이용하여 새로운 사례에 대한 예측결과를 생성하는 전통적인 인공지능기법 중 하나다. 이러한 사례기반추론이 최근 적용이 쉽고 간단하다는 장점과 모형의 갱신이 실시간으로 이루어진다는 점 등으로 인해, 온라인 환경에서의 고객관계관리를 위한 도구로 학계와 실무에서 주목을 받고 있다 하지만, 전통적인 사례기반추론의 경우, 타 인공지능기법에 비해 정확도가 상대적으로 크게 떨어진다는 점이 종종 문제점으로 제기되어 왔다. 이에, 본 연구에서는 사례기반추론의 성과를 획기적으로 개선하기 위한 방법으로 유전자 알고리즘을 활용한 사례기반추론의 동시 최적화 모형을 제안하고자 한다. 본 연구가 제안하는 모형에서는 기존 연구에서 사례기반추론의 성과에 중대한 영향을 미치는 요소들로 제시된 바 있는 사례 특징변수의 상대적 가중치 선정(feature weighting)과 참조사례 선정(instance selection)을 유전자 알고리즘을 이용해 최적화함으로서, 사례간 유사도를 보다 정밀하게 도출하는 동시에 추론의 결과를 왜곡할 수 있는 오류사례의 영향을 최소화하고자 하였다. 제안모형의 유용성을 검증하기 위해, 본 연구에서는 국내 한 전문 인터넷 쇼핑몰의 구매예측모형 구축사례에 제안모형을 적용하여 그 성과를 살펴보았다. 그 결과, 제안모형이 지금까지 기존 연구에서 제안된 다른 사례기반추론 개선모형들은 물론, 로지스틱 회귀분석(LOGIT), 다중판별분석(MDA), 인공신경망(ANN), SVM 등 다른 인공지능 기법들에 비해서도 상대적으로 우수한 성과를 도출할 수 있음을 확인할 수 있었다.

• Steel and Composite Structures
• /
• v.25 no.6
• /
• pp.663-670
• /
• 2017

The problem of layup optimization of the composite laminates involves a very complex multidimensional solution space which is usually non-exhaustively explored using different heuristic computational methods such as genetic algorithms (GA). To ensure the convergence to the global optimum of the applied heuristic during the optimization process it is necessary to evaluate a lot of layup configurations. As a consequence the analysis of an individual layup configuration should be fast enough to maintain the convergence time range to an acceptable level. On the other hand the mechanical behavior analysis of composite laminates for any geometry and boundary condition is very convoluted and is performed by computational expensive numerical tools such as finite element analysis (FEA). In this respect some studies propose very fast FEA models used in layup optimization. However, the lower bound of the execution time of FEA models is determined by the global linear system solving which in some complex applications can be unacceptable. Moreover, in some situation it may be highly preferred to decrease the optimization time with the cost of a small reduction in the analysis accuracy. In this paper we explore some machine learning techniques in order to estimate the failure of a layup configuration. The estimated response can be qualitative (the configuration fails or not) or quantitative (the value of the failure factor). The procedure consists of generating a population of random observations (configurations) spread across solution space and evaluating using a FEA model. The machine learning method is then trained using this population and the trained model is then used to estimate failure in the optimization process. The results obtained are very promising as illustrated with an example where the misclassification rate of the qualitative response is smaller than 2%.

• Asian-Australasian Journal of Animal Sciences
• /
• v.31 no.12
• /
• pp.1980-1990
• /
• 2018

Objective: The aim of this study was to discover the functional impact of non-synonymous single nucleotide polymorphisms (nsSNPs) that were found in selective sweep regions of the Landrace genome Methods: Whole-genome re-sequencing data were obtained from 40 pigs, including 14 Landrace, 16 Yorkshire, and 10 wild boars, which were generated with the Illumina HiSeq 2000 platform. The nsSNPs in the selective sweep regions of the Landrace genome were identified, and the impacts of these variations on protein function were predicted to reveal their potential association with traits of the Landrace breed, such as reproductive capacity. Results: Total of 53,998 nsSNPs in the mapped regions of pigs were identified, and among them, 345 nsSNPs were found in the selective sweep regions of the Landrace genome which were reported previously. The genes featuring these nsSNPs fell into various functional categories, such as reproductive capacity or growth and development during the perinatal period. The impacts of amino acid sequence changes by nsSNPs on protein function were predicted using two in silico SNP prediction algorithms, i.e., sorting intolerant from tolerant and polymorphism phenotyping v2, to reveal their potential roles in biological processes that might be associated with the reproductive capacity of the Landrace breed. Conclusion: The findings elucidated the domestication history of the Landrace breed and illustrated how Landrace domestication led to patterns of genetic variation related to superior reproductive capacity. Our novel findings will help understand the process of Landrace domestication at the genome level and provide SNPs that are informative for breeding.

• Journal of IKEEE
• /
• v.23 no.2
• /
• pp.529-533
• /
• 2019

Combined cycle power plants are often used to produce power. These days prediction of power plant output based on operating parameters is a major concern. This paper presents an approach to using computational intelligence technique to predict the output power of combined cycle power plant. Computational intelligence techniques have been developed and applied to many real world problems. In this paper, tree architectures of fuzzy neural networks are considered to predict the output power. Tree architectures of fuzzy neural networks have an advantage of reducing the number of rules by selecting fuzzy neurons as nodes and relevant inputs as leaves optimally. For the optimization of the networks, two-step optimization method is used. Genetic algorithms optimize the binary structure of the networks by selecting the nodes and leaves as binary, and followed by random signal-based learning further refines the optimized binary connections in the unit interval. To verify the effectiveness of the proposed method, combined cycle power plant dataset obtained from the UCI Machine Learning Repository Database is considered.

• Journal of Industrial Convergence
• /
• v.16 no.4
• /
• pp.33-46
• /
• 2018

Personal credit scoring is an effective tool for banks to properly guide decision profitably on granting loans. Recently, many classification algorithms and models are used in personal credit scoring. Personal credit scoring technology is usually divided into statistical method and non-statistical method. Statistical method includes linear regression, discriminate analysis, logistic regression, and decision tree, etc. Non-statistical method includes linear programming, neural network, genetic algorithm and support vector machine, etc. But for the development of the credit scoring model, there is no consistent conclusion to be drawn regarding which method is the best. In this paper, we will compare the performance of the most common scoring techniques such as logistic regression, neural network, and support vector machines using personal credit data of the financial institution in China. Specifically, we build three models respectively, classify the customers and compare analysis results. According to the results, support vector machine has better performance than logistic regression and neural networks.

• Bulletin of the Korean Chemical Society
• /
• v.21 no.3
• /
• pp.281-290
• /
• 2000

We varied recombination method of fenetic algorithm (GA), i.e., crossover step, to compare efficiency of these methods, and to find more optimum GA method. In one method (A), we select two conformations(parents) to be recombined by systematic combination of lowest energy conformations, and in the other (B), we select them in a ratio proportional to the energy of the conformation. Second variation lies in how to select crossover point. First, we select it randomly(1). Second, we select range of residues where internal energy of the molecule does not vary for more than two residues, select randomly among such regions, and we select either thr first (2a) or the second residue (2b) from the N-terminal side, or the first (2c) or the second residue (2d) from the C-terminal side in the selected region for crossover point. Third, we select longest such hregion, and select such residue(as cases 2) (3a, 3b, 3c or 3d) of the region. These methods were tested in a 2-dimensionl lattice system for 8 different sequences (the same ones used by Unger and Moult., 1993). Results show that compared to Unger and Moult's result(UM) which corresponds to B-1 case, our B-1 case performed similarly in overall. There are many cases where our new methods performed better than UM for some different sequences. When cooling factor affecting higher energy conformation to be accepted in Monte Carlo step was reduced, our B-1 and other cases performed better than UM; we found lower energy conformers, and found same energy conformers in a smaller steps. We discuss importance of cooling factor variation in Monte Carlo simulations of protein folding for different proteins. (A) method tends to find the minimum conformer faster than (B) method, and (3) method is superior or at least equal to (1) method.