Incremental Learning with

 

Learn++ .NSE

© 2001
R. Polikar

Signal Processing  & Pattern Recognition Laboratory

 

===================

Acknowledgement

====================

 

The material described on this page and in the paper is supported by

 

 

 

 

 

 

 

 

 National Science Foundation 

Electrical, Communications and Cyber Systems (ECCS) Division

 Energy, Power and Adaptive Systems (EPAS) Subdivision 

through the CAREER program, under grant number ECS 0239090

and  through Adaptive Intelligent Systems program under grant number ECS 0926159.

 

The material provided below is based upon work supported by the National Science Foundation under
Grant No ECS-0239090, “CAREER: An Ensemble of Classifiers Based Approach for Incremental Learning”. And Grant No ECS-0926159, Incremental Learning from Unbalanced Data in Nonstationary Environments”

 

Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

———————————————————–—————-————————————————————————————

 

For any questions, comments or suggestions, please contact the following e-mail address

This page includes supplementary information for the paper titled:

“Incremental Learning of Concept Drift in Nonstationary Environments,”

by Ryan Elwell and Robi Polikar.

 

 

Data Files:

The following data sets are used in the paper. Each of the following files is a zip file, which includes the data in a comma separated value (CSV) file, as well as a readme.txt file that provides additional information about the dataset (such as features, naming conventions, number of features, training / testing instances, etc.

 

Gaussian Dataset

SEA Dataset

Rotating Checkerboard Datasets—This group includes several datasets with different drift rate scenarios

Weather Dataset—Includes only the preprocessed Offutt Air Force Base in Bellevue, Nebraska dataset used in the paper. For other locations and raw data, see: ftp://ftp.ncdc.noaa.gov/pub/data/gsod/ and the readme.txt file in that directory.


Movie Files:

The following movies show how the environment changes in the Gaussian and the Checkerboard datasets

 

Gaussian Data—This dataset starts with three classes, later adds a new class and removes a class, each of which drift independently.

 

 

 

 

 

 

 

 

 

 

 

 

Checkerboard Data

Constant Drift Rate

Exponentially Increasing Drift Rate

Sinusoidally Changing Drift Rate

Pulsing Drift Rate

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Complete time-averaged performance results for all algorithms, including  NB on SEA and AdaBoost weighting

Rotating checkerboard data showing different snapshots based on the rotation parameter, alpha.

Different drift rate scenarios used in generating variable drift rate for rotating checkerboard dataset

 

Gaussian

SEA

Weather

Bayes

88.1

+/-

0.0

 

 

 

 

 

 

L++.NSE (NB)

84.0

+/-

0.5

96.6

+/-

0.2

75.9

+/-

0.7

Single (NB)

82.3

+/-

1.2

94.7

+/-

0.6

69.4

+/-

1.4

DWM (NB)

84.8

+/-

0.4

96.6

+/-

0.6

71.3

+/-

1.8

SEA (NB)

83.3

+/-

0.3

95.4

+/-

0.4

72.1

+/-

0.8

Adaboost (NB)

82.3

+/-

0.6

93.2

+/-

0.4

72.0

+/-

0.6

L++.NSE (SVM)

81.0

+/-

0.9

96.8

+/-

0.2

78.8

+/-

1.0

Single (SVM)

74.6

+/-

2.4

95.6

+/-

0.4

67.8

+/-

2.0

SEA (SVM)

81.3

+/-

0.5

95.7

+/-

0.2

77.8

+/-

1.1

Adaboost (SVM)

69.1

+/-

1.8

93.2

+/-

0.2

75.2

+/-

0.9

L++.NSE (CART)

82.8

+/-

0.7

95.8

+/-

0.5

75.7

+/-

1.1

Single (CART)

77.7

+/-

1.9

86.7

+/-

1.0

66.8

+/-

2.0

SEA (CART)

81.7

+/-

0.5

95.6

+/-

0.3

72.8

+/-

1.0

Adaboost (CART)

81.4

+/-

0.8

93.3

+/-

0.5

73.2

+/-

0.8

 

CB (constant)

CB (pulse)

CB (exp)

CB (sinusoid)

L++.NSE (NB)

69.9

+/-

1.3

70.5

+/-

1.6

69.1

+/-

1.4

71.1

+/-

1.5

Single (NB)

56.6

+/-

1.7

54.3

+/-

1.7

56.5

+/-

1.7

55.3

+/-

1.7

DWM (NB)

59.6

+/-

1.6

56.4

+/-

1.7

59.6

+/-

1.7

57.9

+/-

1.7

SEA (NB)

60.1

+/-

1.2

64.2

+/-

1.6

61.0

+/-

1.3

63.7

+/-

1.4

Adaboost (NB)

59.9

+/-

1.7

59.0

+/-

1.8

59.9

+/-

1.6

59.2

+/-

1.7

L++.NSE (SVM)

81.9

+/-

0.9

84.0

+/-

0.7

81.6

+/-

0.9

83.5

+/-

0.9

Single (SVM)

76.6

+/-

1.5

79.9

+/-

1.5

76.6

+/-

1.5

78.6

+/-

1.5

SEA (SVM)

71.6

+/-

0.7

78.5

+/-

0.6

73.0

+/-

0.7

75.4

+/-

0.6

Adaboost (SVM)

81.0

+/-

1.0

82.0

+/-

1.1

80.2

+/-

1.1

81.9

+/-

1.1

L++.NSE (CART)

77.3

+/-

1.1

81.2

+/-

1.0

77.0

+/-

1.1

79.4

+/-

1.0

Single (CART)

67.8

+/-

1.9

69.3

+/-

2.6

67.7

+/-

1.9

68.7

+/-

2.3

SEA (CART)

69.3

+/-

0.9

77.2

+/-

0.8

70.6

+/-

0.9

73.2

+/-

0.8

Adaboost (CART

74.7

+/-

1.3

80.5

+/-

1.0

74.8

+/-

1.3

77.4

+/-

1.2