RETRACTED CHAPTER: Local Feature Weighting for Data Classification
Feature weighting is an important task in data analyze, clustering and classification. Traditional algorithms focus on a common weight vector on the whole dataset which can easily lead to sensitiveness to the distribution of data. In contrast, a novel feature weighting algorithm called local feature weighting (LFW) that assign each sample a unique weight vector is proposed in this paper. We use clustering assumption to construct optimization task. Instead of considering the total intra-class and between-class features, we focus on the clustering performance on each training sample and the optimization goals are to minimize the total distances of a training sample to others in the same class and maximize the total distances in different classes. Data weight is added to the target function to emphasis nearby samples and finally use an iterative process to solve our problem. Experiments show that the new algorithm has a good performance on data classification. In addition, we provide a simple version of LFW which has less running time but with little accuracy loss.
KeywordsLocal feature weighting Classification Clustering assumption
- 11.Sugiyama, M.: Local fisher discriminant analysis for supervised dimensionality reduction. In: Proceedings of the 23rd International Conference on Machine Learning, pp. 905–912. ACM, June 2006Google Scholar
- 14.Gilad-Bachrach, R., Navot, A., Tishby, N.: Margin based feature selection-theory and algorithms. In: Proceedings of the Twenty-First International Conference on Machine Learning, p. 43. ACM, July 2004Google Scholar
- 16.Lichman, M.: UCI Machine Learning Repository (2013). http://archive.ics.uci.edu/ml. Irvine, C.A.: University of California, School of Information and Computer Science