Evaluating algorithms that learn from data streams

In the past years, the theory and practice of machine learning and data mining have been focused on static and finite data sets from where learning algorithms generate a static model. In this setting, evaluation metrics and methods are quite well defined. Nowadays, several sources produce data in a stream at high-speed, creating environments with possibly infinite, dynamic and transient data streams. Currently, there is no standard for evaluating algorithms that learn from data streams. In this paper we try to present major issues that bind the evaluation strategy to data stream environments, proposing evaluation methods for both supervised and unsupervised learning.

• Type: Conference
• Year: 2009
• Where: SAC
• Authors: João Gama, Pedro Pereira Rodrigues, Raquel Sebastião
• Keywords: Applied Computing, Data Stream, SAC 2009, Transient Data Streams, finite Data Sets