Presentation is loading. Please wait.

Presentation is loading. Please wait.

On Reducing Classifier Granularity in Mining Concept-Drifting Data Streams Peng Wang, H. Wang, X. Wu, W. Wang, and B. Shi Proc. of the Fifth IEEE International.

Similar presentations


Presentation on theme: "On Reducing Classifier Granularity in Mining Concept-Drifting Data Streams Peng Wang, H. Wang, X. Wu, W. Wang, and B. Shi Proc. of the Fifth IEEE International."— Presentation transcript:

1 On Reducing Classifier Granularity in Mining Concept-Drifting Data Streams Peng Wang, H. Wang, X. Wu, W. Wang, and B. Shi Proc. of the Fifth IEEE International Conference on Data Mining (ICDM ’ 05) Speaker: Yu Jiun Liu Date : 2006/9/26

2 Introduction  State of the art The incrementally updated classifiers. The ensemble classifiers.  Model Granularity Traditional : monolithic This paper : semantic decomposition

3 Motivation  The model is decomposable into smaller components.  The decomposition is semantic-aware in the sense.

4 Monolithic Models  Stream :  Attributes :  Class Label :  Window :  Model (Classifier) : C i

5 Rule-based Models  A rule form :  minsup = 0.3 and minconf = 0.8  Valid rules of W1 are:  Valid rules of W3 are:

6 Algorithm  Phase 1 : Initialization Use the first w records to train all valid rules for window W1. Construct the RS-tree and REC-tree.  Phase 2 : Update When record arrives, insert it into the REC-tree and update the sup. and conf. of the rules matched by it. Delete oldest record and update the value matched by it.

7 Data Structure

8 RS-Tree  A prefix tree with attribute order  Each node N represents a unique rule R : P  Ci  N ’ (P ’  Cj) is a child node of N, iff:

9 REC-Tree  Each record r as a sequence  Node N points to rule in the RS-tree if :

10 Detecting Concept Drifts  percentage V.S. the distribution of the misclassified records. The percentage approach cannot tell us which part of the classifier gives rise to the inaccuracy.

11 Definition

12 Finding Rule Algorithm

13 Update Algorithm

14 Experiments  CPU : 1.7 GHz  Memory : 256MB  Datasets : synthetic and real life dataset. Synthetic :  Real life dataset :  10,344 recodes and 8 dimensions.

15 Effect of model updating  Synthetic  10 dimensions  Window size 5000  4 dimensions changing

16 The relation of concept drifts and

17 Effect of rule composition

18 Accuracy and Time  Window size : 10,000  EC : 10 classifiers, each trained on 1000 records.  Synthetic data.

19 Real life data

20 Conclusion  Overcome the effects of concept drifts.  By reducing granularity, change detection and model update can be more efficient without compromising classification accuracy.


Download ppt "On Reducing Classifier Granularity in Mining Concept-Drifting Data Streams Peng Wang, H. Wang, X. Wu, W. Wang, and B. Shi Proc. of the Fifth IEEE International."

Similar presentations


Ads by Google