I'm having issues seeing the best way to predict a time-series when training on a dataset with different samples.
I have a dataset that shows the weight of 10 rabbits from their first day to their 50th day with 5 measures per days. With it, I also have some different values like the temperature of the room, humidity and other values. There is more but they might really be useful. I would like to predict with the data of a new rabbit, how it will grow. For example, if the rabbit is only 30 days old, that all the other data have been collected so far, I would like to predict its weight for the next 20 days.
I'm not sure how to use all the information for this task. I was thinking of using time-series forecasting methods like ARIMA or LSTM but maybe it wouldn't be the best option with the way I have data on different rabbits. I'm also not sure how to incorporate the multiple dataset for the same feature/target into the model so that I could predict the data of a new rabbit. Using the name of the rabbit as a Categorical variable/feature might work.
Here is a drawing of the data plotted. They oscillate a bit depending of the time of the day and are somewhat similar but with different offset (Can't give those data outside)
So the questions would be :
- What would be the best way to use exploit as much as possible my data to predict the weight of the next 20 days of a new rabbit? (like, for example after 30 days)
- Can I really use the dataset of the 10 different rabbits or should I get only one or an average of them all since they kinda look alike except with somewhat of a different offset? (categorical data seem the way but would it really be ok since I predict a new category?)
- If the other features/values (like the temperature or weather data) are not predictable/controllable but might be predicted with some errors would it be advised to use them?
