9
$\begingroup$

I have about 500 variables per patient, each variable has one continous value and is measured at three different time points (after 2 month and after 1 year). With the regression I would like to predict the treatment outcome for new patients.

Is it possible to use SVM regression with such longitudinal data?

$\endgroup$
  • $\begingroup$ Were you able to find any proper answer? $\endgroup$ – Wazaa Mar 9 '16 at 3:00
1
$\begingroup$

Yes, this is possible. Except that in longitudinal data using Fisher Kernel works better than RBF or Linear ones. A similar setting like that of yours is given in this NIPS paper: http://research.microsoft.com/pubs/147234/NIPS08.pdf

$\endgroup$
1
+50
$\begingroup$

This is an interesting question and I did a quick research.

The OP asked about regression for continuous data. But the paper cited by @Vikram only works for classification.

Lu, Z., Kaye, J., & Leen, T. K. (2009). Hierarchical Fisher Kernels for Longitudinal Data. In Advances in Neural Information Processing Systems.

A related paper for regression I found is the following. Technical details can be found in Section 2.3.

Seok, K. H., Shim, J., Cho, D., Noh, G. J., & Hwang, C. (2011). Semiparametric mixed-effect least squares support vector machine for analyzing pharmacokinetic and pharmacodynamic data. Neurocomputing, 74(17), 3412-3419.

No public software is found but the authors claimed the ease of use at the end of the paper.

The main advantage of the proposed LS-SVM ... is that regression estimators can be easily computed by softwares solving a simple linear equation system. This makes it easier to apply the proposed approach to the analysis of repeated measurement data in practice.

To elaborate a bit more, there are two approaches for regression analysis using SVM (support vector machine):

  • support vector regression (SVR) [Drucker, Harris; Burges, Christopher J. C.; Kaufman, Linda; Smola, Alexander J.; and Vapnik, Vladimir N. (1997); "Support Vector Regression Machines", in Advances in Neural Information Processing Systems 9, NIPS 1996, 155–161]
  • least squares support vector machine (LS-SVM) [Suykens, Johan A. K.; Vandewalle, Joos P. L.; Least squares support vector machine classifiers, Neural Processing Letters, vol. 9, no. 3, Jun. 1999, pp. 293–300.]

The aforementioned Seol et al. (2011) adopted the LS-VSM approach.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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