원문정보
초록
영어
The traditional k-prototypes algorithm is well versed in clustering data with mixed numeric and categorical attributes, while it is limited to complete data. In order to handle incomplete data set with missing values, an improved k-prototypes algorithm is proposed in this paper, which employs a new dissimilarity measure for incomplete data set with mixed numeric and categorical attributes and a new approach to select k objects as the initial prototypes based on the nearest neighbors. The improved k-prototypes algorithm can not only cluster incomplete data with no need to impute the missing values, but also avoid randomness in choosing initial prototypes. To illustrate the accuracy of the established algorithm, traditional k-prototypes algorithm and k-prototypes employing the new dissimilarity measure are compared to the improved k-prototypes algorithm by using data from UCI machine learning repository. The experimental results show that the improved k-prototypes algorithm is superior to the other two algorithms with higher clustering accuracy.
목차
1. Introduction
2. Materials and Methods
2.1. Data Sets
2.2. Problem Description
2.3. Incomplete Set Mixed Dissimilarity (ISMD)
2.4. Improved Selection of Initial k Centers based on Nearest Neighbors
2.5. Improved k-prototypes Algorithm for Incomplete Data with Mixed Attributes
3. Numerical Results
3.1. Evaluation Indexes
3.2. Experimental Results
4. Conclusions and Discussion
Acknowledgements
References