Font Size: a A A

K Nearest Neighbors Algorithm Based Multi-label Classification

Posted on:2012-10-27Degree:MasterType:Thesis
Country:ChinaCandidate:W T ZhangFull Text:PDF
GTID:2218330338457249Subject:Computer technology
Abstract/Summary:PDF Full Text Request
Multi-label classification is a developing field of machine learning, Many methods have been proposed to solve multi-label classification problem, although the problem has just been introduced into machine learning and data mining. These methods can be roughly categorized into two groups:1) Problem Transformation, and 2) Algorithm Transformation. The KNN Methods in this paper focus on the Application of the Multi-label Classification methods, and contains some experiments with comments.The multi-label datasets will have the following characteristics:every pair of instances similar in the feature set, and they will similar in the label set, for the datasets have more than one label, or even more. Based on the characteristics mentioned above, we find that any two samples similar in the label set, will show similarities in the feature set.Based on the assumption that the weighted attribute method of adjusting the weights:we analysis each feature of the feature set, compare the result of classification that ignore the feature mentioned above,with the result of K nearest neighbor based on the label set. We choose the Rank-Loss to measure the differences. When the result is bigger than the threshold, then increases the weight of the feature.This paper focus on the description of experiments, the comparison between the experimental data reflects the past, methods and advantages and disadvantages of this method. the weight adjustment method in this paper, played a certain effect, enhanced the Hamming-Loss in particular.
Keywords/Search Tags:Weight Adjustment, Attribute Weight, Threshold, Algorithm Transformation
PDF Full Text Request
Related items