• Journal of Computing Science and Engineering
• /
• v.8 no.4
• /
• pp.207-214
• /
• 2014

In pedestrian detection applications, one of the most popular frameworks that has received extensive attention in recent years is widely known as a 'Hough forest' (HF). To improve the accuracy of detection, this paper proposes a novel split function to exploit the statistical information of the training set stored in each node during the construction of the forest. The proposed split function makes the trees in the forest more robust to noise and illumination changes. Moreover, the errors of each stage in the training forest are minimized using a global loss function to support trees to track harder training samples. After having the forest trained, the standard HF detector follows up to search for and localize instances in the image. Experimental results showed that the detection performance of the proposed framework was improved significantly with respect to the standard HF and alternating decision forest (ADF) in some public datasets.

• Journal of KIISE:Computer Systems and Theory
• /
• v.34 no.5_6
• /
• pp.195-201
• /
• 2007

A minimum terminal path decomposition of a tree is defined as a partition of the tree into edge-disjoint terminal-to-terminal paths that minimizes the weight of the longest path. In this paper, we present an $O({\mid}V{\mid}^2$time algorithm to find a minimum terminal path decomposition of trees. The algorithm reduces the given optimization problem to the binary search using the corresponding decision problem, the problem to decide whether the cost of a minimum terminal path decomposition is at most l. This decision problem is solved by dynamic programing in a single traversal of the tree.

• Journal of Information Technology Applications and Management
• /
• v.14 no.3
• /
• pp.137-149
• /
• 2007

Most of hospitals, especially small-scale hospitals, have tried to get customers through the Internet as what companies have done recently. There are various attempts that increase visits to one's web-site in plastic surgery hospitals. However, in plastic surgery, there have been few studies on which an attribute contributes to increase the number of web-site visit. In order to derive the important attributes on the number of visit, we compared functional attributes of 30 high-visit plastic surgery web-sites with those of 30 low-visit web-sites using statistical and data mining methods. For analysis, three methods have conducted including Multiple Discriminant Analysis (statistical method), Decision Trees (data mining method), and Artificial Neural Network (data mining method). Furthermore, results of each method have been evaluated one another. The result of this study shows that a few attributes like 'Simulating cyber plastic surgery program', 'recommendation of information' explain the number of the visitors between high and low visit web-site. The methodology employed in this study provides an efficient way of improving satisfaction of visitors of plastic surgery website.

• Journal of KIISE:Software and Applications
• /
• v.32 no.9
• /
• pp.936-947
• /
• 2005

In dependency parsing of long sentences with fewer subjects than predicates, it is difficult to recognize which predicate governs which subject. To handle such syntactic ambiguity between subjects and predicates, this paper proposes an 'S-clause' segmentation method, where an S(ubject)-clause is defined as a group of words containing several predicates and their common subject. We propose an automatic S -clause segmentation method using decision trees. The S-clause information was shown to be very effective in analyzing long sentences, with an improved parsing performance of 5 percent. In addition, the performance in detecting the governor of subjects was improved by $32\%$.

• Journal of KIISE:Software and Applications
• /
• v.34 no.1
• /
• pp.40-51
• /
• 2007

To improve academic achievement, the most students not only participate in regular classes but also take various extra programs such as private lessons, private institutes, and educational TV programs. In this paper, we propose a data mining approach to identify which studying methods or usual life patterns during a school vacation affect changes in the academic achievement. First, we derive various studying methods and life patterns that are thought to be affecting changes in the academic achievement during a school vacation. Second, we propose the method of transforming and analyzing data to apply them to decision trees and association rules, which are representative data mining techniques. Third, we construct decision trees and find association rules from the real survey data of middle school students. We have discovered four representative results from the decision trees. First, for students in the higher rank, there is a tendency that private institutes give a positive effect on the academic achievement. Second, for the most students, the Internet teaming sites nay give a negative effect on the achievement. Third, private lessons that have thought to be making a large impact to the achievement, however, do not make a positive effect on the achievement. Fourth, taking several studying methods in parallel nay give a negative effect on the achievement. In association rules, however, we cannot find any meaningful relationships between academic achievement and usual life patterns during a school vacation. We believe that our approach will be very helpful for teachers and parents to give a good direction both in preparing a studying plan and in selecting studying methods during a school vacation.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1998.06a
• /
• pp.122-127
• /
• 1998

This paper suggests an integrated method for discovering knowledge from a large database. Our approach applies an attribute-oriented concept hierarchy ascension technique to extract generalized data from actural data in databases, induction of decision trees to measure the value of information, and knowledge reduction of rough set theory to remove dispensable attributes and attribute values. The integrated algorithm first reduce the size of database for the concept generalization, reduces the number of attributes by way of elimination condition attributes which have little influence on decision attribute, and finally induces simplified decision rules removing the dispensable attribute values by analyzing the dependency relationships among the attributes.

• Communications for Statistical Applications and Methods
• /
• v.10 no.3
• /
• pp.971-980
• /
• 2003

The tree method can be extended to multivariate responses, such as repeated measure and longitudinal data, by modifying the split function so as to accommodate multiple responses. Recently, some decision trees for multiple responses have been constructed by Segal (1992) and Zhang (1998). Segal suggested a tree can analyze continuous longitudinal response using Mahalanobis distance for within node homogeneity measures and Zhang suggested a tree can analyze multiple binary responses using generalized entropy criterion which is proportional to maximum likelihood of joint distribution of multiple binary responses. In this paper, we will modify CART procedure and suggest a new tree-based method that can analyze multiple binary responses using similarity measures.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.6 no.6
• /
• pp.849-854
• /
• 2011

To minimize the spreading effect from the events of the system, a rule-based expert system is very effective. However, because the events of the large-scale system are diverse and the load condition is very variable, it is very difficult to construct the rule-based expert system. To solve this problem, this paper studies a methodology which constructs a rule-based expert system by applying a CART(Classification and Regression Trees) algorithm based decision tree determination method to event case examples.

• Asian-Australasian Journal of Animal Sciences
• /
• v.15 no.4
• /
• pp.591-595
• /
• 2002

Three hundred thirty five progeny testing steers of Korean beef cattle were evaluated ultrasonically for back fat thickness (BFT), longissimus muscle area (LMA) and intramuscular fat (IF) before slaughter. Class measurements associated with the Korean yield grade and quality grade were also obtained. Residual standard deviation between ultrasonic estimates and carcass measurements of BFT, LMA were 1.49 mm and $0.96cm^2$. The linear correlation coefficients (p<0.01) between ultrasonic estimates and carcass measurements of BFT, LMA and IF were 0.75, 0.57 and 0.67, respectively. Results for improving predictions of yield grade by four methods-the Korean yield grade index equation, fat depth alone, regression and decision tree methods were 75.4%, 79.6%, 64.3% and 81.4%, respectively. We conclude that the decision tree method can easily predict yield grade and is also useful for increasing prediction accuracy rate.

• 한국데이터정보과학회:학술대회논문집
• /
• 2004.04a
• /
• pp.39-50
• /
• 2004

The decision tree approach is most useful in classification problems and to divide the search space into rectangular regions. Decision tree algorithms are used extensively for data mining in many domains such as retail target marketing, fraud dection, data reduction and variable screening, category merging, etc. CHAID(Chi-square Automatic Interaction Detector) uses the chi-squired statistic to determine splitting and is an exploratory method used to study the relationship between a dependent variable and a series of predictor variables. In this paper we propose CHAID algorithm by cube-based proportional sampling and explore CHAID algorithm in view of accuracy and speed by the number of variables.