Feature Selection and Parameter Optimization based on Improved EPSO for Support Vector Machines
碩士 === 國立中興大學 === 資訊管理學系所 === 100 === Feature selection is widely used in many applications in machine learning area. It is used to reduce unnecessary data to improve computing efficiency, and especially important for classification process. The purpose of classification is to build a classifica...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | zh-TW |
| Published: |
2012
|
| Online Access: | http://ndltd.ncl.edu.tw/handle/79225041681679097325 |
| Summary: | 碩士 === 國立中興大學 === 資訊管理學系所 === 100 === Feature selection is widely used in many applications in machine learning area. It is used to reduce unnecessary data to improve computing efficiency, and especially important for classification process. The purpose of classification is to build a classification model and classify data effectively and also help machine to make decisions. Support vector machine (SVM) is usually used to do the classification job.
We proposed a wrapper method with Endocrine based Particle Swarm Optimization and SVM to deal with feature selection and parameter optimization. Hormone regulation mechanism of Artificial Endocrine System can avoid the shock situation in late computing period and local optimal situation. We used datasets from UCI database to evaluate the performance and compared with traditional PSO+SVM scheme. Results showed that our method could avoid local optimal effectively. Besides increasing classification accuracy, our method can decrease the number of features significantly on high-dimensional datasets in limited time.
|
|---|