I am trying to use an LSTM to predict daily usage for users. I have data for 30 days.
Based on business knowledge I know users divide roughly into different categories. E.g. daily users would have a non-zero usage almost every day, weekly users have one or two days of non-zero usage every 7 days and monthly users might have a couple of days with non-zero usage per 30 days.
Samples where every column is one day and each row is access usage for one user.
User 1: 50, 80, 33, 19, 30, 15, ...
User 2: 0, 21, 13, 30, 0, 5, 0, 0, 55, 28, 0, 19, 0, ...
User 3: 11, 2, 11, 56, .....
From above, User1,3 maybe a daily user, User2 maybe a weekly user.
If I only can get info such as user name and access usage.
Can a single lstm model capture this different types of patterns of users?
The goal is to predict the daily usage for next 10 days of each users.
Why I ask is because I tried 100 epochs with learning rate 0.0001, but error still failed.
The prediction result always look like the same even give another input of user name.