On Demand Classification of Data Streams

1
On Demand Classification of Data Streams

• Charu C. Aggarwal
• Jiawei Han
• Philip S. Yu

Proc. 2004 Int. Conf. on Knowledge Discovery and
Data Mining (KDD'04), Seattle, WA, Aug. 2004
Speaker Pei-Min Chou Date2005/04/01
2
Outline

• Introduction
• Supervised Micro-cluster
• Snapshot
• Maintenance Supervised Micro-cluster
• Training Data Stream
• Classification on Demand
• Empirical Results

3
Introduction

• Advances in data storage often grow without limit
  referred to as data streams
• one-pass mining model
• does not recognize the changes and it is too
  expensive to keep track of the entire history
• static classification model likely to drop when
  there is a sudden burst
• Our model
• simultaneous training and testing streams
  used for dynamic classification of data sets

4
Supervised Micro-cluster Modify Micro-cluster

• Only from training data and each with same class
• Data streams
• Multi-dimensional points with
  time stamps T1, Tk .
• Each point contains d dimensions, i.e.,
• A micro-cluster for n points is defined as a (2d
  4) tuple

• - the sum of the squares of the data values
• - the sum of the data values
• - the sum of the squares of the time stamps
• - the sum of the time stamps
• the number of data points
• -variable corresponding to class id corresponds
  to
• the class label of that micro-cluster

5
Snapshot

• not too expensive to keep track history
• storing the behavior of the micro-clusters at
  different moments in time
• if (t mod 2i) 0 but (t mod 2i1)! 0
• reaches max capacity, the oldest snapshot in this
  frame is removed
• geometric time frame
• vary from 0 to a value no larger than log2(T),
• T is the maximum length of the stream
• maximum number
• (max capacity)log2(T)

6
Maintenance Supervised Micro-clusters

• Nearest neighbor and k-means algorithms
• The initial micro-clusters is offline process
• offline ---answers various user queries based on
  the stored summary statistics
• When a new data point Xik arrives, it is either
  added to a micro-cluster, or a new micro-cluster
  is created

7
Classification on Demand

• Construct
• Find the correct time-horizon
• The value of kfit
• Large or small horizon be chosen
• Test

8
Find the correct time-horizon

• Macro-clusters are created over a user-specified
  time horizon h
• Let S(tc) the snapshot of micro-clusters at time
  tc
• S(tc-h) the snapshot of micro-clusters
  at time tc-h
• The new set of micro-clusters N(tc-h) are created
  by subtracting S(tc-h) from S(tc)
• Subtractive property
• Let C1 and C2 be two sets of points such that
• Then

9
Training Data Stream

• A small portion of the stream is used for the
  process of horizon fitting stream segment
• kfit the number of points in the data used and
  the value small as 1 of the data
• remaining portion of the training stream is used
  for the creation and maintenance of the
  class-specific micro-clusters

10
The value of kfit

• Horizon determined classification accuracy
• Process executed periodically for changes
• kfit should be small enough so that the points in
  it reflect the immediate locality of tc
• Qfit pre-specified number of time units
• a part of the training stream
• the class labels are known a-priori
• Nearest neighbor procedure (XeQfit)
• Find the closest micro-cluster in N(tc,h) to X
• compare the class label and true label

11
Large or small horizon be chosen

• The accuracy of all the time horizons which are
  tracked by the geometric time frame are
  determined
• The p time horizons which provide the greatest
  dynamic classification accuracy
• by
• First sight ---smallest
• Stable stream ---large

12
Test

• test stream is a separate process which is
  executed continuously throughout the algorithm
• Insert Xt , nearest neighbor classication process
  is applied using each (Xt belong H)
• results in the determination class lable
• these p class labels reported as the relevant
  class

13
Empirical Results

• Pentium III,512MB,WinXP
• Both real and synthetic
• Advantage
• much higher classification accuracy
• Good scalability in terms of dimensionality and
  the number of class labels
• stable processing rate
• Space-efficient

14
Experiment
15
Experiment
16
Experiment