Time Series Prediction for multi-input multi-output

Question

Muhammad Faisal 2018-1-30

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/379798-time-series-prediction-for-multi-input-multi-output

Assume a simple and hypothetical scenario (similar to my problem) like

ID  VisitDate  Weight  Height  LowBP  HighBP
   Jan 1, 2010  76  5  76  119
 Mar 10, 2010   77  5  73  119
 July 1, 2010   76  5  76  120
 Feb 2, 2009   55   5.5  70  132
 Mar 5, 2009   60  5.5  70  132
 Aug 2, 2009   57  5.5  71  130
….

I would like to predict LowBP and HighBP after 1 month, 2 months, etc.

As you can see the baseline for each patient will be different and the interval is not equal as well. In addition to that I’ve less data and also some missing data.

My questions are:

How to tackle this problem especially the initial stage of preparing the data. Do I need to create some more columns for t+1 values of LowBP and HighBP and copy the next record of each patient as a future value.? In that case the last record will have no future value and I may need to delete it. This is I believe window or lag method.

Do I need to make it a time series? If yes then how it can be multi-input and multi-output?

Do I need to train separately for each ID or I can combine all together? In the real scenario each ID has different slope. It can happen that for same input we have different output for 2 different IDs maybe because of Age or any other factor that is not included as predictors. I've few IDs and for each ID you can say around 4 records on average. So if I try to train NN for each ID, it may not be a good fit.

How could I split for train and test data? Do I need to keep all the last records of each patient as test data. Is this different than the normal supervised or unsupervised prediction where the order doesn't matter.

Which network would be better for this scenario? NarxNet, LSTM, or something else?

I'm using Matlab 2016 for Windows.

Thanks.