Novel Hybrid Decision Support Disease Diagnosis Systems Using Machine Learning Algorithms

摘要

The amount of patient's data in the healthcare facilities are increasing rapidly for the past few decades.The challenge is how to analyze available patient's data to extract relevant knowledge from it and act upon it,in a timely manner.Efficient data mining tools must be utilized to turn the data into knowledge which can aid developing decision-based expert systems that will assist physicians in the early diagnosis of lethal diseases.Such expert systems can reduce human-made errors and mistakes(due to fatigue and tiredness of doctors and practitioners),cost,and wait time.
　　New statistical analysis and data mining techniques are utilized by researchers to develop tools that help healthcare professionals to easily and efficiently diagnose lethal diseases at early stages.It also involves many challenges such as unnecessary attributes(data dimensionality problem),missing features values and best attribute selection to obtain highest diagnostic accuracy.Current research ascertains the problem statement,provides an analysis of existing systems and proposes a series of novel and successful approaches to improve the accuracy of the systems.
　　The first proposed approach in the current study,combines Information Gain method with weighed based pre-processed kNN and Adaptive Neuro-Fuzzy Inference System for the diagnosis of thyroid diseases.This approach selects attributes from a pool of attributes(provided by UCI machine learning repository for thyroid diseases)using Information Gain method.Then applying kNN to handle missing feature values in the selected dataset and fed the preprocessed data to ANFIS in the last stage.The second method presents a twostage approach using Information Gain method and Adaptive Neuro-Fuzzy Inference System for the hepatitis disease diagnosis.The third and last approach of this thesis combines Linear Discriminant Analysis(LDA)with weighed based pre-processed k-Nearest Neighbor(kNN)and Adaptive Neuro-Fuzzy Inference System and applied to thyroid diseases.The obtained results prove that our proposed models are more successful than the previously used diagnostic techniques and can be used as a promising tool for another lethal diseases also.The datasets for the proposed approaches were obtained from University of California Irvin's(UCI)Machine Learning Repository.

目录

声明

DEDICATION

PUBLICATIONS

ACKNOWLEDGEMENTS

Abstract

Table of Contents

List of Figures and Tables

List of Abbreviations／Acronyms

Chapter 1 Introduction

1．1．Overview

1．1．1．Hepatitis Diseases

1．1．2．Thyroid Disease

1．2．Problem Statement

1．3．Research Objectives

1．3．1．General Objective

1．3．2．Specific Objectives

1．3．3．Research Question

1．4．Justification of The Proposed Approaches

1．5．Scope of The Proposed Approaches

1．6．Summary

Chapter 2 Literature Rewew

2．1．Overview(Datamining；Definition and Concept)

2．2．Data Mining as a Process

2．3．Machine Learning Techniques

2．3．1．Naive Bayes Classifier(NBC)

2．3．2．Decision Tree

2．3．3．Artificial Neural Network(ANN)

2．4．Classification Methods

2．4．1．Adaptive Neuro-Fuzzy Inference System(ANFISl)

2．5．Feature Extraction techniques

2．5．1．Principal Component Analysis(PCA)

2．5．2．Symmetrical Uncertainty(SU)

2．5．3．Relief

2．5．5．Focus

2．5．6．Las Vegas Filter

2．5．7．Information Gain

2．5．8．Linear Diseriminant Analysis

2．6．Handling Incomplete Data

2．6．1．Weighed based pre-processing using kNN algorithm

2．7．Metrics

2．8．Diagnosis History of Experimented Diseases

2．8．1．Hepatitis Disease Diagnosis Background

2．8．2．Thyroid Disease Diagnosis Background

2．9．Summary

Chapter 3 Materials and Methods

3．1．Overview

3．2．Data mining

3．2．1．Data Retrieval

3．2．2．Feature Extraction

3．2．3．Data Preprocessing

3．2．4．Applying Proposed Approaches

3．2．5．Evaluation

3．2．6．Results

3．3．Summary

Chapter 4 Experimental resub and Discussion

4．1．Overview

4．2．Presentation of Data

4．3．Results Diseussion

4．3．1．IG-KNN-ANFIS Results

4．3．2．IG-ANFIS Results

4．3．3．LDA-KNN-ANFIS Results

4．4．Summary

Chapter 5 Conclusion and Future Work

5．1．Overview

5．2．conclusion

5．3．Future Work

References