I am using STAI-YI scores in a study to classify anxiety levels as low, medium and high. However, does the individual STAI score has any significance over just a range? Is trying to generate the exact STAI score of subjects using artificial neural networks an overkill?
With most psychopathology-related scales, it's best to use exact scores rather than ranges. When you reduce continuous scores to dichotomous variables, you can muddle the interpretability and, in many cases, reduce your statistical power.
Not sure how this applies exactly in the case of estimating scores from neural networks, but I assume you can just predict a continuous variable instead of dichotomized range variables. This seems more meaningful to me.
I have found papers suggesting cut-offs for making the problem binary, e.g. 39-40, while other papers suggest that the cut-off is higher for geriatric population (e.g. 44-45). So depending on the population the cut-offs may vary significantly, especially if you need three classes instead of two I was using.
However I agree with Scott's answer above, you can modify the problem for predicting a continuous value instead of a categorical one, and use regression instead of classification. You can do it with neural networks as well.
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Publication