# How to begin reading about data mining?

I'm a novice who is going to start reading about data mining. I have basic knowledge of AI and statistics. Since many say that machine learning also plays an important role in data mining, is it necessary to read about machine learning before I could go on with data mining?

Being somewhat in this position myself, I'll try to give some insight.

Firstly, download the Elements of Statistical Learning. It presumes calculus and linear algebra, and although it is very technical, it is also extremely well written.

Secondly (or firstly) look at Andrew Ng's tutorials on machine learning.

Thirdly, get some data, and start attempting to analyse data. You'll need to split into training and test sets, and then build models on the training set and test them against the test set. I found the caret package for R very useful for all of this. After that its practice, practice practice (like almost everything else).

• you'll scare the poor man away forever! Commented Aug 27, 2011 at 19:43
• Andew Ng's course will be 'offered free and online' to student's world wide during fall 2011 according to ml-class.org Commented Aug 27, 2011 at 20:41

Introduction to Data Mining by Tan, Steinbech, Kumar is the best intro book out there

http://www.amazon.com/Introduction-Data-Mining-Pang-Ning-Tan/dp/0321321367

save EoSL for when you want to dig deeper. It's more of a reference.

Data mining can be descriptive or predictive.

On the one hand, if you are interested in descriptive data mining, then machine learning won't help.

On the other hand, if you are interested in predictive data mining, then machine learning will help you understand that you try to minimize the unknown risk (expectation of the loss function) when minimizing the empirical risk: you will keep in mind overfitting, generalization error and cross-validation. For instance, for a matter of consistency, the $k$-NN for a training sample of size $n$ should be such that:

• $k$ goes to infinity when $n$ goes to infinity,
• $\frac{k}{n}$ goes to 0 when $n$ goes to infinity.
• It's worth noting that some authors like to make a distinction between DM and ML depending on the magnitude of $k/n$. I personally like Radford Neale's approach, in his course on Statistical Methods for Machine Learning and Data Mining: Many machine learning problems have a large number of variables, Data mining applications often involve very large numbers of cases.
– chl
Commented Jul 11, 2011 at 22:37

I only add another very good source of tutorials on data mining/machine learning by Tom Mitchell.

He explains it very clearly and You can also download his presentations from his website (together with watching his lectures there).