Using additionnal features for time series problem?

How can i use other features to forecast use DNN?

You can have as many features as you consider necessary. Just make sure that you pre-process them appropriately. For example, if you add a feature called “color” and you have 3 possible colors, you should one-hot encode this feature into 3 features (actually 2 will do but lets say 3 for ease of understanding).

Sometimes the feature has an inherit order, for example “Review” with the options (Bad, Ok, Excellent). In this case, instead of one-hot encoding you may want to preprocess it as 1, 2, 3 where 1 = bad, 2 = ok, 3 = Excellent. The increased order of the values will give a hint to the model to the order of importance.

Another step you will need is to normalize the data. This is, try to normalize all numerical values so they end up between 0 and 1 or -1 to 1.

These are just some basic concepts and ideas. There’s more to it, but hopefully this sheds light on your question.