Development of multi-label classification algorithm based on correlations among labels

Dissertant

Diyab, Raid Hasan Salih

Thesis advisor

Thabtah, Fadi Fayiz Abd al-Jabir

University

Philadelphia University

Faculty

Faculty of Information Technology

Department

Department of Computer Science

University Country

Jordan

Degree

Master

Degree Date

2013

English Abstract

-Multi label classification is concerned with learning from set of instances that are associated with a set of labels, that is, an instance could be associated with multiple labels at the same time.

This task occurs frequently in application areas like text categorization, multimedia classification, bioinformatics, protein function classification and semantic scene classification.

Current multi-label classification methods could be divided into two parts.

The first part is called problem transformation methods, which transform multi-label classification problem into single label classification problem, and then apply any single label classifier to solve the problem.

The second part is called algorithm adaptation methods, which adapt an existing single label classification algorithm to handle multi-label data.

The following are some of the research challenges in the field of multi-label classification problem: 1.

Design a hierarchical structure for multi- label to manage label correlationships.

2.

To extract relevant label sets from multi-label data set.

3.

A novel approach that uses both problem transformation methods, and algorithm adaptation methods, to improve performance and accuracy for multi-label classifier.

In this thesis, we propose a multi-label classification algorithm based on correlations among labels, that uses both problem transformation methods and algorithm adaptation method.

The algorithm begins with transforming multi-label dataset into single label dataset using least frequent label criteria, and then applies PART algorithm on the transformed dataset.

Also the algorithm tries to get benefit from positive correlations among labels using predictive Apriori algorithm.

The output of the algorithm is multilabels rules.

The algorithm has been evaluated using two multi-label datasets ( "Emotions"," Yeast") and three evaluation measures (Accuracy, Hamming Loss, Harmonic Mean).

Further, we show by experiments that this algorithm has a fair accuracy comparing with other related algorithms.

Main Subjects

Mathematics

Topics

• Algorithms
• Data mining

No. of Pages

61

Table of Contents

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-546361