You can also clone this example project to get started easily.
What is new in ELKI 0.7.0? Too much, see the release notes, please!
What is ELKI exactly?
ELKI is a Java based data mining toolkit. We focus on cluster analysis and outlier detection, because there are plenty of tools available for classification already. But there is a kNN classifier, and a number of frequent itemset mining algorithms in ELKI, too.
ELKI is highly modular. You can combine almost everything with almost everything else. In particular, you can combine algorithms such as DBSCAN, with arbitrary distance functions, and you can choose from many index structures to accelerate the algorithm. But because we separate them well, you can add a new index, or a new distance function, or a new data type, and still benefit from the other parts. In other tools such as R, you cannot easily add a new distance function into an arbitrary algorithm and get good performance - all the fast code in R is written in C and Fortran; and cannot be easily extended this way. In ELKI, you can define a new data type, new distance function, new index, and still use most algorithms. (Some algorithms may have prerequisites that e.g. your new data type does not fulfill, of course).
ELKI is also very fast. Of course a good C code can be faster - but then it usually is not as modular and easy to extend anymore.
ELKI is documented. We have JavaDoc, and we annotate classes with their scientific references (see a list of all references we have). So you know which algorithm a class is supposed to implement, and can look up details there. This makes it very useful for science.
ELKI is not: a turnkey solution. It aims at researchers, developers and data scientists. If you have a SQL database, and want to do a point-and-click analysis of your data, please get a business solution instead with commercial support.