Feature Reduction for Multi Label Classification of Discrete Data

We describe a novel multi-label classification algorithm which works for discrete data. A matrix which gives the membership value of each discrete value of each attribute for every class. For a test pattern, looking at the values taken by each attribute, we find the subset of classes to which the pattern belongs. If the number of classes are large or the number of features are large, the space and time complexity of this algorithm will go up. To mitigate this problems, we have carried out feature selection before carrying out classification. We have compared two feature reduction techniques for getting good results. The results have been compared with the algorithm multi-label KNN or ML-KNN and found to give good results. Using feature reduction our classification accuracy and running time for algorithm is improved. The performance of the above algorithm is evaluated using some benchmark datasets and the results have been compared with the algorithm multi-label KNN or ML-KNN and found to give good results