基于不确定性建模的数据挖掘

1 Introduction

1.1 Types of Uncertainty

1.2 Uncertainty Modeling and Data Mining

1.3 Related Works

References

2 Induction and Learning

2.1 Introduction

2.2 Machine Learning

2.2.1 Searching in Hypothesis Space

2.2.2 Supervised Learning

2.2.3 Unsupervised Leaming

2.2.4 Instance-Based Learning

2.3 Data Mining and Algorithms

2.3.1 Why Do We Need Data Mining?

2.3.2 How Do We do Data Mining?

2.3.3 Artificial Neural Networks

2.3.4 Support Vector Machines

2.4 Measurement of Classifiers

2.4.1 ROC Analysis for Classification

2.4.2 Area Under the ROC Curve

2.5 Summary

References

3 Label Semantics Theory

3.1 Uncertainty Modeling with Labels

3.1.1 Fuzzy Logic

3.1.2 Computing with Words

3.1.3 Mass Assignment Theory

3.2 Label Semantics

3.2.1 Epistemic View of Label Semantics

3.2.2 Random Set Framework

3.2.3 Appropriateness Degrees

3.2.4 Assumptions for Data Analysis

3.2.5 Linguistic Translation

3.3 Fuzzy Discretization

3.3.1 Percentile-Based Discretization

3.3.2 Entropy-Based Discretization

3.4 Reasoning with Fuzzy Labels

3.4.1 Conditional Distribution Given Mass Assignments

3.4.2 Logical Expressions of Fuzzy Labels

3.4.3 Linguistic Interpretation of Appropriate Labels

3.4.4 Evidence Theory and Mass Assignment

3.5 Label Relations

3.6 Summary

References

4 Linguistic Decision Trees for Classification

4.1 Introduction

4.2 Tree Induction

4.2.1 Entropy

4.2.2 Soft Decision Trees

4.3 Linguistic Decision for Classification

4.3.1 Branch Probability

4.3.2 Classification by LDT

4.3.3 Linguistic ID3 Algorithm

4.4 Experimental Studies

4.4.1 Influence of the Threshold

4.4.2 Overlapping Between Fuzzy Labels

4.5 Comparison Studies

4.6 Merging of Branches

4.6.1 Forward Merging Algorithm

4.6.2 Dual-Branch LDTs

4.6.3 Experimental Studies for Forward Merging

4.6.4 ROC Analysis for Forward Merging

4.7 Linguistic Reasoning

4.7.1 Linguistic Interpretation of an LDT

4.7.2 Linguistic Constraints

4.7.3 Classification of Fuzzy Data

4.8 Summary

References

5 Linguistic Decision Trees for Prediction

5.1 Prediction Trees

5.2 Linguistic Prediction Trees

5.2.1 Branch Evaluation

5.2.2 Defuzzification

5.2.3 Linguistic ID3 Algorithm for Prediction

5.2.4 Forward Branch Merging for Prediction

5.3 Experimental Studies

5.3.1 3D Surface Regression

5.3.2 Abalone and Boston Housing Problem

5.3.3 Prediction of Sunspots

5.3.4 Flood Forecasting

5.4 Query Evaluation

5.4.1 Single Queries

5.4.2 Compound Queries

5.5 ROC Analysis for Prediction

5.5.1 Predictors and Probabilistic Classifiers

5.5.2 AUC Value for Prediction

5.6 Summary

References

6 Bayesian Methods Based on Label Semantics

6.1 Introduction

6.2 Naive Bayes

6.2.1 Bayes Theorem

6.2.2 Fuzzy Naive Bayes

6.3 Fuzzy Semi-Naive Bayes

6.4 Online Fuzzy Bayesian Prediction

6.4.1 Bayesian Methods

6.4.2 Online Learning

6.5 Bayesian Estimation Trees

6.5.1 Bayesian Estimation Given an LDT

6.5.2 Bayesian Estimation from a Set of Trees

6.6 Experimental Studies

6.7 Summary

References

7 Unsupervised Learning with Label Semantics

7.1 Introduction

7.2 Non-Parametric Density Estimation

7.3 Clustering

7.3.1 Logical Distance

7.3.2 Clustering of Mixed Objects

7.4 Experimental Studies

7.4.1 Logical Distance Example

7.4.2 Images and Labels Clustering

7.5 Summary

References

8 Linguistic FOIL and Multiple Attribute Hierarchy for Decision Making

8.1 Introduction

8.2 Rule Induction

8.3 Multi-Dimensional Label Semantics

8.4 Linguistic FOIL

8.4.1 Information Heuristics for LFOIL

8.4.2 Linguistic Rule Generation

8.4.3 Class Probabilities Given a Rule Base

8.5 Experimental Studies

8.6 Multiple Attribute Decision Making

8.6.1 Linguistic Attribute Hierarchies

8.6.2 Information Propagation Using LDT

8.7 Summary

References

9 A Prototype Theory Interpretation of Label Semantics

9.1 Introduction

9.2 Prototype Semantics for Vague Concepts

9.2.1 Uncertainty Measures about the Similarity Neighborhoods Determined by Vague Concepts

9.2.2 Relating Prototype Theory and Label Semantics

9.2.3 Gaussian-Type Density Function

9.3 Vague Information Coarsening in Theory of Prototypes

9.4 Linguistic Inference Systems

9.5 Summary

References

10 Prototype Theory for Learning

10.1 Introduction

10.1.1 General Rule Induction Process

10.1.2 A Clustering Based Rule Coarsening

10.2 Linguistic Modeling of Time Series Predictions

10.2.1 Mackey-Glass Time Series Prediction

10.2.2 Prediction of Sunspots

10.3 Summary

References

11 Prototype-Based Rule Systems

11.1 Introduction

11.2 Prototype-Based IF-THEN Rules

11.3 Rule Induction Based on Data Clustering and Least-Square Regression

11.4 Rule Learning Using a Conjugate Gradient Algorithm

11.5 Applications in Prediction Problems

11.5.1 Surface Predication

11.5.2 Mackey-Glass Time Series Prediction

11.5.3 Prediction of Sunspots

11.6 Summary

References

12 Information Cells and Information Cell Mixture Models

12.1 Introduction

12.2 Information Cell for Cognitive Representation of Vague Concept Semantics

12.3 Information Cell Mixture Model (ICMM) for Semantic Representation of Complex Concept

12.4 Learning Information Cell Mixture Model from Data Set

12.4.1 Objective Function Based on Positive Density Function..

12.4.2 Updating Probability Distribution of Information Cells...

12.4.3 Updating Density Functions of Information Cells

12.4.4 Information Cell Updating Algorithm

12.4.5 Learning Component Number of ICMM

12.5 Experimental Study

12.6 Summary

References

Machine learning and data mining are inseparably connected with uncertainty. The observable data for learning is usually imprecise,incomplete or noisy. Uncertainty Modeling for Data Mining A Label Semantics Approach introduces 'label semantics', a fuzzy-logic-based theory for modeling uncertainty. Several new data mining algorithms based on label semantics are proposed and tested on real-world datasets. A prototype interpretation of label semantics and new prototype-based data mining algorithms are also discussed. This book offers a valuable resource for postgraduates, researchers and other professionals in the fields of data mining, fuzzy computing anduncertainty reasoning.

《基于不确定性建模的数据挖掘（英文版）》首先从基础开始介绍了Label Semantics的理论体系，并基于该理论基础提出了一系列新的数据挖掘算法。然后，对每个算法进行了深入详细的研究，并和其他常用算法做了定量的比较。最后，介绍了最新的Prototype理论对label semantics的解释。