Customer Reviews

Automating the Design of Data Mining Algorithms: An Evolutionary Computation Approach (Natural Computing Series)

5 star:		(0)
4 star:		(1)
3 star:		(0)
2 star:		(0)
1 star:		(0)

 

 

This book covers data mining via automatic, evolutionary generation of rule induction algorithms using context based process grammar. It is relatively short at around 200 pages, the presence of equations and algorithms is relatively sparse but the ones which are included are useful and clearly presented.

The real strength of this book in my opinion is the very clear and straightforward way the authors explain the fundamental problems associated with building such a system, and the pros and cons of various methods for solving them (ie how to create an efficient yet accurate fitness function, considerations for designing a process grammar, avoiding overfitting etc).

I have read a number of other data mining books on similar subjects, many of them are quick to jump in to 'this is how you implement our method' without in depth explanation of the pros and cons of various aspects or implicit assumptions inherent in the technique. Either that or they approach the subject from an overly theoretical, mathematical point of view, focusing on proving the optimality of their technique in some limited scenario while ignoring the fact that most real world (high dimensional) datasets are too complex to be mined in this way. In contrast, I felt this book struck a good balance between theory and application.

Overall this is a good, solid introduction to data mining via generative rule induction. I would have given it 5 stars if it included a few more types of algorithms or made available sample code.

I would recommend it highly to beginners interested in this topic. Those who already have experience with similar techniques such as generative rule induction via neuro-fuzzy methods may or may not find it useful.