Incremental Learning with Learn++ .NSE
Robi Polikar
This page includes supplementary information for the paper titled:
Incremental Learning of Concept Drift in Nonstationary Environments,
by Ryan Elwell and Robi Polikar.
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.
Constant Drift Rate
Exponentially Increasing Drift Rate
Sinusoidally Changing Drift Rate
Pulsing Drift Rate
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
Complete time-averaged performance results for all algorithms, including NB on SEA and AdaBoost weighting
| 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 |
| 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 |
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.
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: