Customer Reviews

Neural, Novel & Hybrid Algorithms for Time Series Prediction

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

 

 

Prediction methods for time series are a multi-million dollar industry and are of upmost importance in financial engineering, weather prediction, logistics, network modeling, and myriads of other fields. This book gives an overview of various methodologies for time series prediction, and is written for readers with substantial experience in this area. The author emphasizes that time series prediction is more of an art rather than a science, with the practitioner usually employing hybrids of established techniques, only some of which have a rigorous mathematical foundation. In fact, despite the subject matter, this book is very lean on mathematics, and the reader will have to consult other books for a more detailed mathematical treatment. The NPredict package accompanying the book is designed to run on an NT and a DOS platform, and illustrates the main points in the book. Readers who have familiarity with the authors earlier books on neural networks will definitely find this one easier to follow. It is, again, not written for beginning students, but the author does a fairly good job of presenting the material for the advanced reader.

I usually don't pay attention to readers's comments or reviews.
Because after all this is a very subjective matter. I bought the book for two reasons neither of which has to do with a reader's review. The first reason was because I have a copy of Timothy's book on Practical Neural Networks in C++, which I found excellent, and the second reason was because I had previewed chapter one before I bought the book and liked it very much for what it had to say and the way it said it. Timothy's books are for a wide audience of intelligent people, not necessarily all rocket scientists, and although a mathematician himself, restricts math as much as possible so people do not get bogged down by the math and loose the forest for the trees. On the other hand there is sufficient amount of bibliography for any one who is interested to pursue most rigorous or more exotic approaches. The code examples are good and the executable file NPREDICT, works without any further processing, for those who don't want to mess with code and compiling. The treatment of Box-Jekins ARMA model,and the multivariate example on temperature and precipitation is very good. The book is highly recomended to any one who has little or no knowledge of the subject, and wants to understand what time series is all about

An excellant practical, hands-on book for filtering, pre-processing data for forecasting using ARMA, NN and combination of both systems. It contains a lot of background tools to do the job. NN discusstion was short since I would like to see more discussion on the different network types (but you might check his other NN with C++ books). The book also discusses confidence intervals with ARMA and NN. The majority of the book is on filtering, pre-processing etc... It just depends on what type of book you are looking for. The description contains an error. The book doesnt contain any FUZZY techniques. Masters is an excellant author.

Probably my impression would be better if author setup right expectation for this book. As a mathematically inclined person I was disappointed by fact that author left all explanation of crucial concepts and algorithms behind the scenes just referencing "It is described in other books" - see subjects about Maximum Entropy Methods, details in ARIMA. Reader is left to take some concepts for granted without clear understanding of subject. Meantime author missed some important subjects in Neural Networks and even digital filters like recursive digital filters which proved to be superior to filters described in the book. The rest of theory is quite heuristic and based on unproved concepts and author's "feelings" that is not acceptable at least for me. Sometimes author refers to the program code to explain methods but the quality of code whish to be better. He states about code that "It is wildly extravagant in its memory usage in order to save a small amount of execution time. This reflects modern hardware characteristics". Unfortunately this concept is wrong because modern software design requires simple and clear code rather then weird code with questionable improvement in performance, which is difficult to use and read. Shortly, I would not recommend to buy or use this book because it might be only useful for beginning programmers who does not care about the subject to code.

This book delivers what the title promises. If the reader is interested in learning advanced preprocessing techniques this book will equip them with(some)powerful tools. The code alone in the accompanying disk is worth many times the cover price.

Not much on neural nets. A good overview of a signals and systems textbook for those who want to learn about filters without all the math. I was disappointed that there wasn't a results section for the NPREDICT tool, just a bunch of flags and parameter garbage to tweak.

Having an Engineering degree and having read many many technical books, I found the level of explanation and content including examples to be very poor. I bought the book on the belief (Aus$100) that I could learn something from it after having read the synopsis on amazons site. Please Amazon be more accurate with the content analsys of the books you are selling.I know I have been ripped and there is not much I can do about that! I am just warning other people to be wary about purchasing books that they do not know the content of. Trial and error can be so expensive if you live in Australia!

Do not buy this book unless you already read it in a library and just like it to be on the shelf. All ideas are very well known and code is almost useless.
I found much better sources for free on the web.