A New Feature Engineering Approach for Multi-label Classification using Feature Encoding and Soft-loss

Abstract

This thesis aims to improve the performance of multi-label classification (MLC) potentially. The research objectives are to improve the MLC performance using feature encoding and Soft-loss. This work attempts to drive three research questions and investigate scientific approaches to respond to the questions to achieve the research objectives. The thesis's contribution is divided into three folds : (i) Results of comparing state-of-the-art MLC methods with the non-communicable disease dataset. (ii) Feature reconstruction technique using an AutoEncoder network that encodes the features and labels, which improves the efficiency of MLC on the standard dataset. (iii) Applying the label patterns of the data to improve the classification performance.

-