Hi I have a large data set of objects, each containing a set of the same attributes. The attributes are measured quantities like height, width, etc. The data is arranged in a time series so that the value for an attribute for an object is indexed by its time. I want to create a multivariate model of the attributes but was not sure if I should look into parametric or non-parametric methods. Most of the attributes are correlated and dependent on one another so when creating a model I need to have variables representing the other attributes in some way. One of my ultimate goals is to perform time series data mining for intervention analysis. Should I be using hierarchical models since I have multiple objects? Can this possibly be a survival model using a time-to-failure distribution? Any suggestions for methods to look at would be helpful. An example of what my data looks like is shown below. Thanks 
$\begingroup$
$\endgroup$
4
-
$\begingroup$ There is no information here on what the "attributes" are. $\endgroup$– Nick CoxCommented Jul 8, 2015 at 17:46
-
$\begingroup$ Statistically, they can each come from different families of distributions. $\endgroup$– cavsCommented Jul 8, 2015 at 19:28
-
1$\begingroup$ Still no information! Do you mean something qualitative, counted, measured? $\endgroup$– Nick CoxCommented Jul 8, 2015 at 19:45
-
$\begingroup$ Sorry, they are all measure quantities like height, width, etc. $\endgroup$– cavsCommented Jul 8, 2015 at 20:01
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
0
If your objective is Intervention Detection then you definitely want to use parametric methods. When you construct an ARIMA model it is possible to detect pulses , level shifts , seasonal pulses and local time trends using procedures suggested by Tsay and others as described here.
Take a look at the flow diagram in http://www.autobox.com/cms/index.php/blog/entry/build-or-make-your-own-arima-forecasting-model to give you some guidance.
