Partially supervised clustering algorithms for pattern recognition

Dissertant

Labzour, Nadiyah Tazi

Thesis advisor

Bin Said, Amini

University

Al Akhawayn University

Faculty

School of Science and Engineering

Department

Computer Science

University Country

Morocco

Degree

Master

Degree Date

1998

English Abstract

This thesis describes a large class of models called semi-supervised clustering.

Algorithms in this category are (i) clustering algorithms that (ii) use a finite design set of labeled data to {Hi) help clustering algorithms partition a finite set of unlabeled data and then (iv) terminate without the capability to label other points.

First, we show that the semi-supervised point-prototype clustering algorithm (ssPPC) as described in [5] can produce degenerate partitions of the unlabeled data set.

We propose two alternative approaches that guarantee non-degenerate classes.

We apply the improved algorithms to Iris data set and show that their performance is superior to the IDS decision tree and Quick propagation neural networks.

Then, we apply the general partially supervised clustering approach to agglomerative hierarchical clustering (AHC) algorithms used with relational data, we call this procedure semi-supervised agglomerative hierarchical clustering (ssAHC) algorithm, we show through experimentation that the partially-supervised clustering approach helps AHC algorithms improve their results.

Finally, we apply the semi-supervised clustering approach (ssAHC) to text categorization.

We apply ssAHC to Reuters database of documents, and show that its performance is superior to Bayes classifier, and Expectation Maximization algorithm combined with Bayes classifier, we show that the semi-supervised technique helps AHC get better performance.

Main Subjects

Information Technology and Computer Science

Topics

• Sensory perception
• Perception
• Algorithms

No. of Pages

67

Table of Contents

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-647589