Title: Attribute weight gain ratio: new distance measure to select optimal features from multivalued attributes

Authors: L.N.C. Prakash K.; Kodali Anuradha

Addresses: Department of Computer Science and Engineering, AITS Rajampet, AP, India ' Department of Computer Science and Engineering, GRIET, Hyderabad, TS, India

Abstract: Identifying the appropriate features or attributes remains the most prominent stage of any information retrieval and knowledge discovery. The process involves selecting specific features and their subsets holding the vital portion of the data. However, despite the prominence of this stage, most feature selection techniques opt for choosing mono-valued features. Accordingly, these techniques cannot be extended to use in multivalued attributes which require capturing different features from the dataset in parallel. To enable optimal feature selection for multivalued attributes, this manuscript proposes a novel technique aiming at calculating the optimal combination of multivalued attribute entries regarding clusters in unsupervised learning, and classes in supervised learning. The proposal is a distance metric that motivated from the traditional relevance assessing metrics information gain (IG) and gains ratio. To analyse the performance of the proposed technique, the classification approach SVM trained on optimal multivalued attribute features selected using proposed distance measuring metric, which is further used to perform classification process. Also, to evince the significance of the proposed distance measuring metric regarding clustering process, k-means clustering method with attribute weight gain ratio (AWGR) is executed on benchmark datasets. Simulation results depict superior performance of the model for feature selection for multivalued attributes.

Keywords: multiclass attributes; optimal feature; k-means clustering; transaction weight; mining techniques.

DOI: 10.1504/IJAIP.2024.139947

International Journal of Advanced Intelligence Paradigms, 2024 Vol.28 No.1/2, pp.110 - 127

Received: 03 Mar 2018
Accepted: 23 May 2018
Published online: 15 Jul 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article