Towards Parameter-free Data Mining: Mining Educational Data with Yacaree

In the educational arena, Data Mining techniques are acquiring a major importance since the appearance of the e-learning environments. These systems log all the activity carried out by students and instructors, and this raw data, adequately processed, may offer useful knowledge about the learning process for instructors. But data mining techniques are out of the reach of most teachers, e.g., for humanities or law studies. Thus, if we want to help users of all disciplines, we need to work out data mining tools that do not require much tuning or technical understanding from the user. In particular, this is relevant for the case of association rules: all the available algorithms up to recent work depend on one or more parameters (confidence, support,etc) whose value is to be set by the user, and whose semantics may not be easy to grasp. Likewise, the number of rules which obtain as output is often large, and most of them are redundant and non-interesting for decision making [Garćıa et al. 2007]. There is, thus, a clear need to design and implement parameter-free data mining algorithms addressed to “non-experts”, and they must stand reasonably well a comparison with other “expert”-oriented algorithms. To the best of our knowledge, yacaree [Balcázar 2011] is the first parameter-free association miner implemented. Here, we compare this system with other three well-known association rule miners: the Apriori [Agrawal and Srikant 1994] and Predictive Apriori [Scheffer 2001] tools from Weka, and Borgelt’s Apriori implementation [Borgelt 2003].