0
$\begingroup$

I have time series data: students that learned in groups for minimum 3 times and maximum 10 times and for each learning group session had to state if they faced a motivational OR a cognitive problem, what strategies they used to regulate their problem, and how satisfied they were overall with their learning.

First, I want to split my sample into four different subgroups:
- a first group includes only data from the 1st measurement and here only people that were rather satisfied with their 1st learning process
- another subgroup only including data from the 1st measurement and here only learners that are rather not satisfied with their 1st learning process
- a third subgroup only including data from the last measurement and here only people that are rather satisfied with their last learning process
- a last subgroup with data only from the last measurement and here only people that are rather not satisfied with their last learning process

Then, I want to plot a decision tree showing the probability with which SATISFIED learners faced a motivational vs. a cognitive challenge during their first learning process, and the strategies they activated here. As well, I want to plot the same decision tree for UNSATISFIED learners for their first learning process.

Further, I would like to draw both trees for the 10th learning session. My plan is it to predict what strategies all learners will use in their 10th learning process (as I said, I have only for like 20% of the participants the data for the 10th learning process). How can I proceed to make such a prediction?
a) Do I need a classification or a regression tree? (each learner could mention more strategies, and could mention also more strategies of a same type)
b) And is it possible to predict what strategies all students would face in their 10th learning process only based on the data for the 1st learning process? Or should I then also include data from all available data points? However, I cannot imagine how to do that.

enter image description here

Improve this question
$\endgroup$

1 Answer 1

Reset to default
2
$\begingroup$

It is neither classification, nor regression, since you don't want to predict anything, you want to compare groups given some known criteria. Decision trees are not used for this kind of tasks. This is a hypothesis testing problem. You may also find interesting the following thread: Distinguishing between two groups in statistics and machine learning: hypothesis test vs. classification vs. clustering.

Improve this answer
$\endgroup$
3
  • $\begingroup$ Thank you so much! I have the same data set between 3 and 10 times per person (time series). What if I only include the data for the first learning process (1. measurement) and then try to predict what problems all learners will face in their last (10th) learning process? Even though I have data for the 10th learning process for around 20% of the participants, there are still 80% left for making those predictions. Would such an analysis work? $\endgroup$
    – Nadine M.
    Commented Oct 9, 2018 at 14:19
  • $\begingroup$ @NadineM. but this is something completely different then you described, isn't it..? $\endgroup$
    – Tim
    Commented Oct 9, 2018 at 14:36
  • $\begingroup$ Okay, I edited the ad now. $\endgroup$
    – Nadine M.
    Commented Oct 9, 2018 at 15:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.