Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer software > Program design,software engineering > Programming > Database theory and systems

The Research of Cover Algorithm Based on Ensemble Learning

Author FengLunKuo
Tutor JiaRuiYu
School Anhui University
Course Computer Software and Theory
Keywords Covering Algorithm Covering clustering Differences Integrated learning Selective Ensemble
CLC TP311.13
Type Master's thesis
Year 2010
Downloads 57
Quotes 0
Download Dissertation

Classification and clustering are two important data mining techniques, classification data set with the same class label data to establish rules or data model, by these rules or model correctly classified. Clustering is no category label data centralized data grouped by the similarity group objects similar to the high degree of similarity between the two groups is low. Constructive Neural Networks is a new type of neural network, it will be divided into a number of independent function modules network functions, the entire network can be layered gradually constructed. Compared with traditional neural networks, Constructive Neural Networks to build relatively simple, easy to understand, relatively independent of the internal function modules, the design is simple, and can be processed in parallel with the large-scale network, solve massive data to solve the traditional neural network structure is complex, the training speed slow, extended neural network applications shows a huge advantage and potential. Constructive Neural network based on covering ideas proposed departure from the geometric meaning of the neuron model, its core field covering algorithm, the algorithm first gradually in the sample set projection domain construct containing only the same type of data, the spherical region \spherical region, then a common class label composed of a unified output. The integrated learning technique is the use of the multiple learning to solve the same problem, and this can significantly improve generalization capability and stability of the learning system. Traditional covering algorithm and can not be achieved on the learning process of the incremental sample paper proposes incremental learning algorithm of coverage based on integrated learning, learning new sample by sample weights set to increase, and the incremental samples for different situations give the corresponding algorithm, achieving a successful learning process covering algorithm for incremental samples. Traditional areas covering algorithm because of \the one hand, the algorithm significantly reduced rejection samples, on the other hand also significantly improve the generalization ability of the algorithm. Covering clustering algorithm is the traditional areas covering algorithm used in the cluster analysis, take advantage of the characteristics of the local aggregation of clustering data clustering algorithm, the algorithm has a fast clustering, and a relatively simple set of parameters, this paper covering clustering The algorithm for the K-means algorithm to explore the initial center, the improved algorithm can not only significantly reduce the number of iterations of the K-means, but also help to identify the K-means clustering effect. For covering clustering algorithm clustering effect is not ideal, this paper covering algorithm itself features proposed based on the the center match new cluster label matching method, and on this basis, covering clustering algorithm based on integrated learning, the The algorithm can improve the the coverage algorithm clustering effect. Covering algorithm classification or clustering result is the number of \and radius to determine, from which the cover classification and clustering algorithms to achieve selective integrated learning center similar diversity measure used to integrate individual learning, the improved algorithm can greatly reduce number.

Related Dissertations
More Dissertations