Knowledge Discovery from data streams is one of the most relevant challenges that we face today. Data mining algorithms for analyzing static data sets, assuming stationary distributions, unlimited memory, and generating static models are almost obsolete for the real challenging problems we are faced nowadays.

Albert's research is not in the established data mining. He does steps ahead, focusing his research in cutting-the-edge topics, methods and algorithms. His work focuses on evolving data, going far way from iid assumption. Models are not static, training sets are not fixed. Albert's approach tries to capture the dynamics of patterns in the problem under study: how patterns evolve, grow, decrease and die.

The book focus on the design of learning algorithms for evolving and time-changing data streams. The basic building block is the adaptive sliding window algorithm, ADWIN, for change detection and value estimation with strong theoretical guarantees. The book presents evidence of the advantages of the framework in predictive learning, clustering, and closed frequent tree mining from time-changing data streams. The work opens new research opportunities and a large space for applications. The work presents an in-depth study in streaming mining. The author carefully presents theoretical analysis of the proposed methods and algorithms. It introduces new contributions on several different aspects of the problem. Overall this is an interesting dissertation that makes significant contributions to mining time-changing streams.

It is not a surprise, the author's good record of publications in top-level conferences. The ideas come from and reflect a world in movement: a pleasure to read. I'm sure you will enjoy it.

Porto, 21 November, 2009

João Gama