0
$\begingroup$

I am trying to predict if students could meet their assignments deadline or not based on days to complete and number of other assignments that they are supposed to complete. I found out that: There is no spearman or pearson correlation (close to 0) between inputs and output. This is clear from the plots as well. I built a binary logistic regression on this data and the model doesn't fit. Since there might be a nonlinear relation between inputs and output, I also built a neural network and error was so high. Could it be concluded that I can't predict the output based on these input variables?

$\endgroup$
6
  • 1
    $\begingroup$ How did you decide that the regression model "doesn't fit"? And likewise, for the neural network, what error did you get and what baseline error did you compare it to? $\endgroup$ Jan 4, 2017 at 17:48
  • $\begingroup$ I have used SPSS, for binary logistic regression Cox,Snell R squared is 0.007 and Negelkerke R squared is 0.01. Also classification table shows that correct percentage is 53. Similar result in classification table is given for NN $\endgroup$
    – Heidi Mon
    Jan 4, 2017 at 19:21
  • $\begingroup$ I assume that by "correct percentage" you mean the percent of cases that were classified correctly. Is that training error or test error? $\endgroup$ Jan 4, 2017 at 22:17
  • $\begingroup$ And what's the base rate? $\endgroup$ Jan 5, 2017 at 2:23
  • $\begingroup$ Is that training error or test error?....both are about 50%...and what do yo mean by base rate? $\endgroup$
    – Heidi Mon
    Jan 5, 2017 at 14:50

1 Answer 1

0
$\begingroup$

My guess is that you're right: you can't accurately predict the dependent variable using these features. Since your models have training errors near the base rate, the features don't seem to have much predictive value. It's possible, but very unlikely, that a predictively useful relationship exists that neither the logistic-regression model nor the neural network was able to capture.

$\endgroup$

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.