Feature Scaling for Online Learning of Binary Classifiers

スポンサーリンク

概要

• 論文の詳細を見る

• Scaling feature values is an important task in numerous machine learning tasks. Often feature scaling is conducted in an unsupervised manner as a preprocessing task prior to learning. By conducting feature scaling at train time, we can quickly adapt to the changes in data stream as well as scale features to optimize classification accuracy. We study the effect of feature scaling at training time for one-pass online binary classification algorithms. We propose a joint supervised feature scaling method to simultaneously scale features during training time. We incorporate the proposed method into a binary logistic regression model and train using stochastic gradient descent in a one-pass online learning setting.

• 2012-07-19

著者

• Hitoshi Iba

  Graduate School Of Information Science And Technology The University Of Tokyo

• Danushka Bollegala

  Graduate School of Information Science and Technology, The University of Tokyo

• Danushka Bollegala

  Graduate School Of Information Science And Technology The University Of Tokyo

関連論文

• A Comparative Study of Similarity Measures for Cross-Domain Sentiment Classification
• Feature Scaling for Online Learning of Binary Classifiers

スポンサーリンク