In the age of big data, predictive analytics is all around us. Probably you have heard about the Netflix Prize, awarding US$1,000,000 to anyone capable of beating their recommendation engine by 10%. The winning algorithm was based on the the combination of many classifiers. Assessing and Improving Prediction and Classification basically explains you how to built an even better algorithm. It is specialized in the classifier assessment and how to combine many classifiers (or train many versions of it).
Let’s be clear, in this field, it’s the best book I know. It is not an intro in predictive analytics, it’s a very advanced book. I learnt a lot. Moreover, it make you think a lot. You will need a pen and paper while reading the book to note all those ideas that pop up. Sometimes, I closed the book and spend some minutes thinking on what I’ve just read and how it can be applied to real cases I encountered. It is therefore also quite motivational.
The algorithms are implemented in a blend of C and C++, which seems somewhat an odd choice those days. I usually dislike the use of code in the books anyway. Nevertheless, in this case, almost every line is documented. It is therefore adding value to the rest of the book.
The book remind me of Data Preparation for Data Mining (a good companion). Both are pushing boundaries of your knowledge to the top level. On the drawback, it’s not easy reading. Illustrations are not really fancies nor frequent. You will also need a lot of concentration. That’s sadly the price to pay for any book of this quality.
My favorite chapter is the gating one. It’s basically how to train a meta-classifier to learn how to combine the weak classifier depending on the context. It’s really pushing predictive analytics to the next level. It make me remembers my PhD years experimenting on various meta-learning methods. This kind of chapter can give you a real advance over the competition regarding predictive analytics accuracy.
In conclusion, if you are serious about predictive analytics, if you want to achieve better results, this book is exactly what you need.
Let's stay in touch with the newsletter
Possible related posts: