0
$\begingroup$

Data - I have one continuous dependent variable and 10-15 factor independent variables. Tool - R

What kind of models I can use other than linear regression and regression trees?

I applied Linear regression in R and I achieved R^2 as 0.3. Regression trees are also not clear.

Are any other algorithms which can help me for this kind of dataset?

$\endgroup$
2
  • 1
    $\begingroup$ There is a plethora of algorithms you can try. Look for example at the package caret. However it is better to think about the data first and then choose the algorithm. There are multiple reasons why you got low $R^2$, this might be a problem with the particular regression specification, not with regression method per se. $\endgroup$
    – mpiktas
    May 6 '15 at 7:50
  • $\begingroup$ See also here. $\endgroup$ May 6 '15 at 9:37
2
$\begingroup$

An $R^2$ of 0.3 is not necessarily a sign of trouble. In most social science circumstances that is a perfectly acceptable number.

When I discuss this with my students I typically start with a model on who marries who. The $R^2$ is typically extremely high (by social science standards): about 0.6. As I present that result the students are typically appalled by the fact that I cannot explain 40% of the variance. If I turn the discussion around and ask them if they could still marry someone they love if we could explain a 100% of the variance, then that 40% unexplained variance becomes reassuring or even a bit low.

$\endgroup$
1
  • $\begingroup$ +1 Inspired by your answer, I found this on Google, which is probably a very lousy movie. $\endgroup$
    – Zhubarb
    May 6 '15 at 7:59
0
$\begingroup$

Your R Square values will be low with these models since both linear regression and regression trees will try to put your data points in different buckets and map all the points in the bucket with the mean value for that bucket.

This may be ok if you are trying to understand the mechanics behind or find key variables etc.

If prediction is your goal I would suggest Boosted Regression Trees (specifically Gradient Boosted Machine) which will be build more complex models but you will lose interpretability.

$\endgroup$

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