Inhalt

•  
   Contents
•  
   List of Figures
•  
   List of Tables
•  
   1 Introduction
•  
   1.1 Application Example
•  
   1.2 Learning from Data Streams
•  
   1.3 Incremental, Adaptive and Evolving Learning
•  
   1.4 Contribution and Outline of the Thesis
•  
   2 Background
•  
   2.1 Machine Learning
•  
   2.2 Supervised Learning from Data Streams
•  
   2.3 Concept Change over Time
•  
   2.4 Change Detection Methods
•  
  2.5 Adaptive Supervised Learning: Related Work
•  
   2.5.1 Rule-based learning
•  
   2.5.2 Decision trees learning
•  
   2.5.3 Instance-based learning
•  
   2.5.4 Ensemble methods
•  
  3 Instance-Based Classification and Regression
•  
   3.1 Instance-Based Learning
•  
   3.1.1 Classification
•  
   3.1.2 Regression
•  
   3.2 Instance-Based versus Model-Based Learning
•  
   3.3 Instance-Based Learning on Data Streams
•  
   3.4 IBLStreams
•  
   3.4.1 Classification
•  
   3.4.2 Regression
•  
   3.4.3 Parameter adaptation in IBLStreams
•  
   3.4.4 Implementation issues
•  
   3.5 Experiments
•  
   3.5.1 IBLStreams versus other instance-based methods
•  
   3.5.2 Evaluating the parameter adaptation schemes
•  
   3.5.3 IBLStreams versus state-of-the-art model-based methods
•  
   3.5.3.1 Classification
•  
   3.5.3.2 Regression
•  
   3.6 Discussion and Conclusion
•  
   4 Evolving Fuzzy Pattern Trees
•  
   4.1 Introduction to Fuzzy Sets
•  
   4.1.1 Operations on Fuzzy Sets
•  
   4.1.2 Aggregation Operations on Fuzzy Sets
•  
  4.2 Data-Driven Fuzzy Modeling
•  
   4.2.1 Fuzzy Subsethood-Based Algorithm
•  
   4.2.2 Fuzzy Decision Trees
•  
   4.3 Fuzzy Pattern Trees
•  
   4.3.1 Bottom-Up Induction of Fuzzy Pattern Trees
•  
   4.3.2 Top-Down Induction of Fuzzy Pattern Trees
•  
   4.4 Evolving Fuzzy Pattern Trees
•  
   4.4.1 Performance Monitoring and Hypothesis Testing
•  
   4.4.2 Summary of the Algorithm
•  
   4.4.3 Refinements on the Neighbor Trees Generation
•  
   4.5 Empirical Evaluation
•  
  4.5.1 Performance Comparison
•  
   4.5.1.1 Synthetic Data
•  
   4.5.1.2 Real Data
•  
   4.5.2 Model Size
•  
   4.5.3 Sensitivity Towards Significance Levels and Operators Retraining
•  
   4.6 Summary and Conclusion
•  
  5 Survival Analysis on Event Streams
•  
   5.1 Introduction
•  
   5.2 Survival Analysis
•  
   5.2.1 Censored data
•  
   5.2.2 Survival Functions
•  
   5.2.3 Estimating the Survival Function
•  
   5.2.4 Prognostic Factors for Survival
•  
   5.3 Survival Analysis on Data Streams
•  
   5.3.1 Left Censoring
•  
   5.3.2 Parallel Event Sequences
•  
   5.3.3 Adaptive ML Estimation
•  
   5.4 Case Study: Earthquake Analysis
•  
   5.4.1 Data Generation
•  
   5.4.2 Results
•  
   5.5 Case Study: Twitter Data
•  
   5.6 Conclusion
•  
  6 Recovery Analysis for Adaptive Learning
•  
   6.1 Introduction
•  
   6.2 Learning under concept drift
•  
   6.3 Recovery Analysis
•  
   6.3.1 Main idea and experimental protocol
•  
   6.3.2 Bounding the optimal generalization performance
•  
   6.3.3 Recovery measures
•  
   6.3.4 Defining pure streams
•  
   6.3.5 Further practical issues
•  
   6.4 A comparison of algorithms
•  
   6.5 Experiments and results
•  
   6.5.1 Binary classification
•  
   6.5.2 Multiclass classification
•  
   6.5.3 Regression
•  
   6.5.4 Recovery measures
•  
   6.5.5 Summary of the experiments
•  
   6.6 Conclusion
•  
  7 Conclusion
•  
   7.1 Original Contributions
•  
   7.2 Future Research
•  
  A Methods
•  
   A.1 Adaptive Hoeffding Tree
•  
   A.2 Adaptive Model Rules
•  
   A.3 Fast Incremental Model Trees with Drift Detection
•  
   A.4 FLEXible Fuzzy Inference Systems
•  
   B MOA
•  
   B.1 Stream Generators
•  
   B.2 Online Evaluation
•  
  C M-Tree
•  
   C.1 Distance Function
•  
  D Data Sets
•  
  D.1 Synthetic Data Sets
•  
   D.1.1 Hyperplane data
•  
   D.1.2 Distance to hyperplane data
•  
   D.1.3 Random trees data
•  
   D.1.4 Radial basis function data
•  
   D.1.5 SEA concept functions
•  
   D.1.6 STAGGER concept functions
•  
  D.2 Synthetic Data Manipulation
•  
   D.2.1 Concept drift simulation
•  
   D.2.2 Sampling drift simulation
•  
  D.3 Real Data Sets
•  
   D.3.1 Cover type data
•  
   D.3.2 Mushroom data
•  
   D.3.3 Page blocks data
•  
   D.3.4 Letter recognition
•  
   D.3.5 StatLog (shuttle) data
•  
   D.3.6 Skin segmentation data
•  
   D.3.7 MAGIC gamma telescope data
•  
   D.3.8 Breast cancer Wisconsin
•  
   D.3.9 Parkinson's telemonitoring data
•  
   D.3.10 Slice localization data
•  
   D.3.11 Bank32h
•  
   D.3.12 Census-house
•  
  D.4 Event Streams
•  
   D.4.1 Earthquake event stream
•  
   D.4.2 Twitter stream
•  
  E Incremental Statistics
•  
   E.1 Incremental Moments
•  
   E.2 Shifting Moments
•  
   Bibliography