| Project/Area Number |
10680392
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Intelligent informatics
|
|---|
| Research Institution | SCIENCE UNIVERSITY OF TOKYO |
|---|
Principal Investigator |
SHIMURA Masamichi SCIENCE UNIVERSITY OF TOKYO, INDUSTRIAL ADMINISTRATION, PROFESSOR, 理工学部, 教授 (30029409)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
OHWADA Hayato SCIENCE UNIVERSITY OF TOKYO, INDUSTRIAL ADMINISTRATION, LECTURER, 理工学部, 講師 (30203954)
|
|---|
| Project Period (FY) |
1998 – 1999
|
| Project Status |
Completed (Fiscal Year 1999)
|
| Budget Amount *help |
¥3,300,000 (Direct Cost: ¥3,300,000)
Fiscal Year 1999: ¥1,600,000 (Direct Cost: ¥1,600,000)
Fiscal Year 1998: ¥1,700,000 (Direct Cost: ¥1,700,000)
|
|---|
| Keywords | data mining / committee machine / pattern classification / symbiotic evolution / learning / discrimination function / neural network / genetic algorism / 線形機械 / 非線形識別関数 |
|---|
| Research Abstract |
The purpose of data mining is to extract useful knowledge embedded in a huge data base. The stochastic analysis for mining is often very difficult when the data is not linear, not dependent, not numeric. The technology of machine learning can be used for data mining, but the knowledge to be acquired in data mining is concerned with characteristics of objects, clustering, classification, prediction etc. In this project, clustering and classification are main target to investigate and a new method to generate decision trees for classification has been developed based on the symbiotic evolution approach which evolves a population of neurons through genetic algorisms. All parameters including crossover probability, mutation rate, population size are investigated, and the method of finding the fitness value to combine the chromesomes of the best-performing neurons has been developed. The fitness value is generally a very important parameter but is difficult to represent, since the tree is required to be as simple as possible and also to provide correct classification. Our experimental results show the proposed method is useful for construction of simple and correct trees comparing with the existing methods. Furthermore the symbiotic evolution has been applied to a committee machine to classify patterns. Generally patterns which are not linearly separable are difficult to be classified by linear discrimination functions, since the hill-climbing method for learning cannot converge to find the optimal weights. In our system the weight coefficients are obtained by using the symbiotic evolution algorism. Our system is favorable compared with other classifiers.
|
