Density-based Logistic Regression

Density-based Logistic Regression (DLR) is a novel and general approach to nonlinear classification. It has the potential of achieving the following:

The main idea is to map the data to a feature space based on kernel smoothing. A discriminative model is then learned to optimize the feature weights as well as the bandwidth of a Nadaraya-Watson kernel density estimator. For further information, please see our KDD'13 paper [1]. Currently, the DLR suite has two algorithms

Matlab Code Download

For now, only binary classification is available. Multi-way classification is still under testing but will be released soon.

DLR-x code (paper [1])

DLR-b code (recommended!)(technical report [2]) (This report is intended to be submitted to a conference or journal in the near future. For now, you should cite this technical report along with the original DLR paper.)

Reference

[1]. W. Chen, Y. Chen, Y. Mao, and B. Guo, Density-Based Logistic Regression, Proc. ACM SIGKDD Conference (KDD-13), 2013.

[2]. W. Chen, Y. Chen, DLR-b: Density-based Logistic Regression with bins for Large-scale Nonlinear Learning, WUSTL Technical Report, 2013.

Acknowledgement

This research is supported in part by NSF grants CNS-1017701, CCF-1215302, and a Microsoft Research New Faculty Fellowship.

For any questions or bug reports, please contact Yixin Chen (wenlinchen AT wustl.edu) or Yixin Chen (chen AT cse.wustl.edu).