Feature-weighted logistic regression, ridge regression, SVM and SVR

Feature-weighted logistic regression, ridge regression, SVM and SVR

In classification and regression problems, sometimes you know the importance of the features in prior. This situation makes it necessary to assign the features different weights when training a prediction model. Here we modify the objective function of 4 popular classification/regression algorithms and provide the feature-weighted versions of logistic regression, ridge regression, SVM and SVR, all for linear cases. The SVM and SVR are based on our modified liblinear toolbox, which is included in this package. Note that we actually assign different penalty weights to the features, so larger penalty indicates smaller importance.
Please refer to Ke Yan et al., Improving the transfer ability of prediction models for electronic noses, Sensors and Actuators B: Chemical, 2015