Students at risk of dropping out - minimum course time

Students at risk of dropping out - minimum course time

by Andrzej Gałuszka -
Number of replies: 7

Hi Moodlers!

I want test Moodle Machine Learning.  I prepare 10 student account's and test course. Time between course start time and course end time is 8 minutes. 5 accounts going on the route to pass, 5 accounts only complete 1 of 4 activities. After 2 test's (and more) Moodle always predict: all student's account's is in student at risk of dropping out. I think that result's is because course time is too short. I have right? How is minimum course time to received good result's? Maybe i doing another mistake?

Average of ratings: -
In reply to Andrzej Gałuszka

Re: Students at risk of dropping out - minimum course time

by Andrzej Gałuszka -

i attach screenshot how it look

Attachment moodleorg.png
Attachment moodleorg3.png
Average of ratings: -
In reply to Andrzej Gałuszka

Re: Students at risk of dropping out - minimum course time

by David Monllaó -

Hi Piotr,

I don't know which time-splitting method you use but the 8 minutes course duration may be a problem yes. E.g. if you use 'quarters' time-splitting method the course duration is split in 4 parts and the ML backend will try to match the activity in each of the 4 quarters with the students labels. Using short time frames like 8 minutes can be a problem because you may have generated all the activity during the first 2 minutes and the last prediction is the displayed one. If the ML backend is not able to match the activity in the last quarter of the course (the last 2 mins) with the target the predictions will not be good. I would suggest you 'quarters accumulative' time-splitting method as it read all the activity logs from the start of the course.

PD: Please, use the python backend for any serious test smile

Average of ratings: -
In reply to David Monllaó

Re: Students at risk of dropping out - minimum course time

by Elizabeth Dalton -

David, what would you recommend for the shortest time to try to test a simple model and generate training data?

Average of ratings: -
In reply to Elizabeth Dalton

Re: Students at risk of dropping out - minimum course time

by David Monllaó -

Hi Elizabeth,

You could try with "Single range" time-splitting method (it uses data from the beginning of the course to its end) and a few of the existing core indicators. You could use whatever is simpler or whatever you are interested on as 'target'. It is important that you generate activity logs for training according to your target and that you generate similar activity logs in the ongoing course that will receive the predictions so the predictions you get from the ML backend make sense.

I would also suggest to use more than a couple of students as the more samples the ML backends can use for training the better.

Average of ratings: -
In reply to Andrzej Gałuszka

Re: Students at risk of dropping out - minimum course time

by Elizabeth Dalton -

Hi Piotr,

I've been experimenting with trying to test "super short" courses like this, and I think 8 minutes is too short, but it may depend on how you have cron set up. How often are you running cron?

Average of ratings: -
In reply to Elizabeth Dalton

Re: Students at risk of dropping out - minimum course time

by Andrzej Gałuszka -

Hi, i don't running cron. I think clicking "get predictions" work. I set running cron to 5 minutes and time course increased to 20 minutes. Problem not solved after that. (only accuracy set higher)

Average of ratings: -
In reply to Andrzej Gałuszka

Re: Students at risk of dropping out - minimum course time

by Elizabeth Dalton -

Piotr, are the activities set with due dates at different times throughout the course?

Average of ratings: -