I have time series dataset with 7 categorical features (Names) and each of them has 3-5 numerical features (results of activities of each person) for the period 1996-2017 by months. Is it possible to create a model that will be forecasting "Y" on the base of values for selected Names. For example: John + Jack = Y? in 2017-02-01.
Yes, you certainly can. You have to convert all your categorical variables into one numbers. This means, names in your
Name2 columns can be converted using one-hot encoder.
After converting them, you may consider some feature engineering techniques to such dimensionality reduction based on your scenario.
After this, you can use whatever the model is in your mind to solve your problem. Also, please take a peak at this post.