Re: K Nearest Neighbors in time series forecasting

Dnia Thu, 23 Aug 2007 14:38:14 +0100
"David Jones" <dajxxxx@xxxxxxxxx> napisał(a):

I don't understand what you mean by "neighbours" here, if we are considering cross-validation of an AR model, not k-NN.

Does this mean that for the case "times s1 to s2 are being reserved for cross-validation", you would undertake the two tasks:

(1-a) model-fitting : fit on all rows where none of the elements include times s1 to s2;
(1-b) model-fitting : fit on all rows where the RHS (final element) is not in times s1 to s2;

(2-c) assessment : assess on all rows excluding those used in (1-a) ;
(2-d) assessment : assess on all rows for times s1 to s2 (ie all rows excluding those used in (1-b) )

?

See for example http://citeseer.ist.psu.edu/uller97predicting.html . Authors also use
cross-validation for estimating parameters of the model.

Best regards
--
[ Wit Jakuczun <W.Jakuczun [at] wlogsolutions.com> ]
[ WLOG Solutions http://www.wlogsolutions.com ]
.