Using Particle Swarm Method to Optimize the Proportion of Class Label for Prototype Generation in Nearest Neighbor Class
Nearest classification with prototype generation methods would be successful on classification in data mining. In this paper, we modify the encoded form of the individual to combine with the proportion for each class label as the extra attributes in each
- PDF / 1,700,232 Bytes
- 7 Pages / 439.37 x 666.142 pts Page_size
- 44 Downloads / 143 Views
Abstract Nearest classification with prototype generation methods would be successful on classification in data mining. In this paper, we modify the encoded form of the individual to combine with the proportion for each class label as the extra attributes in each individual solution, besides the use of the PSO algorithm with the Pittsburgh’s encoding method that include the attributes of all of the prototypes and get the perfect accuracy, and then to raise up the rate of prediction accuracy. Keywords Particle swarm optimization algorithms Classification
Prototype generation Evolutionary
Introduction The nearest neighbor algorithm [1] has a significant effect on classification prediction. To calculate the similarity between the predicted target and the known samples is the way to find the nearest neighbor. This method provides a very high accuracy rate and having the characteristic that the more precise with the more the number of samples. However, there are some drawbacks for the method. The costs of the calculation are too high and the accuracy is susceptible to noise interference. J.-L. Chen (&) Department of Multimedia Design, Tajen University, Tajen, Taiwan, Republic of China e-mail: [email protected] S.-P. Tseng Department of Computer Science and Information Engineering, National Cheng Kung University, Cheng Kung, Taiwan, Republic of China J.-L. Chen C.-S. Yang The Institute of Computer and Communication Engineering, National Cheng Kung University, Cheng Kung, Taiwan, Republic of China
Y.-M. Huang et al. (eds.), Advanced Technologies, Embedded and Multimedia for Human-centric Computing, Lecture Notes in Electrical Engineering 260, DOI: 10.1007/978-94-007-7262-5_28, Springer Science+Business Media Dordrecht 2014
239
240
J.-L. Chen et al.
In order to solve the above problem, some method can achieve this goal by thinking about reduction of the number of samples. There are two main proposed methods that try to select the reasonable ones from all of the samples and then to perform the nearest neighbor algorithm. These two methods are named prototype selection (PS) [2–7] and the prototype generation (PG) [8–11]. For the purposes of prototype selection to perform the classification, those prototypes are base on the new selection of suitable samples from the training set. There are two main methods for PS problem. One is the concentration method [2], the main idea is to avoid the proportion of certain types of samples are more large than others that make the error decision. For the reason that those samples are eliminated for the properties may be too similar or unrelated. By the second method, the main purpose is to focus on removing those samples would interfere with decision or cause confusion then the follow-up prediction would be more accurate [3]. For prototype generation (PG), not only choose the appropriate samples but also modify the attributes of individual sample. At the result, the decision of classification would be more obvious and distinguished [12, 13]. The main purpose of PG m
Data Loading...
