Educational Performance Analysis

Click to use the system; it has been developed by M. Digas as part of his M.Sc. dissertation (at Hellenic Open University).

You can use the attached data file for experimentation (it is in .arff format).

Make sure you have familiarised yourself with the workflow before actually trying to analyse data.

• First, you need to download the attachment now and then upload it into the system. You can then select which attributes you want to include in the model. You can work with multiple files; all options are available via the Preprocess tab.
• Then you can build and visualize models using the option in the Classify tab (the software either comes by calling WEKA methods or by using GATree).

Following up on that work, V. Hamilou developed a variant of that system as part of her M.Sc. dissertation (Hellenic Open University - to be linked in due course).

The follow-up system is in Greek and it can be tested here.

That variant is primarily aimed at a student who might like to see whether the system will predict pass or failure for a given course module, given the student's homework grades up to that time. Additionally, a tutor can also use the system to see the same prediction for a subset of the students in his/her class.

Here's a brief description of how to use the system:

• As a student, you pick your module, how many homework assignments per year this module has, and for how many of them you have grades on which to base the prediction. You then input the grades and decide whether you want that prediction to be based on a model derived from data based on last year only or on all previous years. The model outputs a prediction with some confidence levels and allows you to browse the training and test data files.
• As a teacher, you do the same but supply students' grades in a file, instead of inputting them one by one (unzip the student data file and use the file based on whether you want a prediction to be made based on two, three or four homework assignemnst).

Watch out: all attached data is real but arbitrarily selected from much larger data files. The predictions are simply meant to convey a look-n-feel of the process.