Technique Recommendation to Recover Missing Labels by Performance Analysis of Multi-Label Machine Learning Algorithms

Sayed Mortaza Kazemi, Zahidullah Sherzad, Muhammad Haleem, and Nasratullah Zarif

Technique Recommendation to Recover Missing Labels by Performance Analysis of Multi-Label Machine Learning Algorithms

Sayed Mortaza Kazemi, Zahidullah Sherzad, Muhammad Haleem, and Nasratullah Zarif

Volume 4 Issue 1 | Dec 2022

DOI: 10.31841/KJET.2022.25

Total Downloads: 5

Download PDF

Abstract

With today’s technological growth Machine learning is revalued as a tool that provides the ability in automating systems. The machines learn from experience based on the
available datasets with no intervention from humans. Here in this research, we consider a variant of the classification, multi-label learning where each instance belongs to more than one label simultaneously. In contrast to other classification tasks, there are a number of challenges, the most significant is to figure out the class when the labels are absent. We compared various multi-label methods from algorithm adaptation and problem transformation perspectives. We applied the algorithms to eight datasets. Problem transformation algorithms are applied to analyze label correlation in a positive sense and then both in a positive and negative sense. We contribute by recommending the technique to address the issue of missing labels by either relying solely on label correlation or by combining it with data locality. Experimental results on multiple benchmark data sets demonstrate that MLLCRS-ML outperforms other cutting-edge techniques.