PDA

View Full Version : Data Mining Techniques



mperez
06-13-2009, 04:56 AM
Hi,I'm doing a study to characterize the students that choose a school. I want to find why they choose this school, why they choose each course of the school. they are going to fill some questionaires and I want to know which is the best DM Technique to analyze the results: Clustering, Association Rules??? (or just only statistics with SPSS - I rather use DM)For a problem like is, what's your adviceThanx
MP

Luykx2
06-13-2009, 05:35 PM
Depends. Do you know the schools/courses the students have picked? If so, then you should probably go for a classification-technique. Either the school, or the course can be the class (or both, if you concatenate the attributes). You could start with using a decision-tree classifier like J48. It's fast and the results are easily interpreted. Same holds for rule-based learners.

If school and/or courses are unknown you could use some clustering techniques. Whether the found clusters will give you insight in which school/course they're going to pick, I cannot say (I doubt it actually).

mperez
06-16-2009, 09:37 AM
Hi,
all the students have chosen the same school and the school has several courses. Therefore, inside a school, I want to characterize the students who have chosen that school and a course of that school.
The information I want should be more descriptive rather than predictive.
I hope you can help me.

Tx
MP

Mark
06-16-2009, 06:54 PM
Hi,

To characterize the school you could use simple statistics, classification or clustering.

Classification based on trees or rules would focus on the characteristics of students that take a particular course. Since there are probably a large number of courses, and students take more than one course, you could build a model for each course in turn using a one-against-the-rest binary class attribute.

Clustering could be used as well. The courses taken by students can be represented by binary attributes. Clusters will hold similar students (i.e. similar according to the courses they take and the other variables you measure).

Cheers,
Mark.

mperez
06-17-2009, 03:44 AM
Each student choose only one course (e.g., medicine,...), not several courses.
What you said (in the last post) still apply to this scenario?

Tx
MP

Mark
06-18-2009, 06:25 PM
Each student choose only one course (e.g., medicine,...), not several courses.
What you said (in the last post) still apply to this scenario?
MP

Yes, using rule or tree learners would find rules that discriminate those who take a particular course from those that don't.

Cheers,
Mark.