I have a time series of crop area that looks like below-
I have other parameters- profit, interest rate, crop prices, and dollar value, and I believe that these parameters can predict the area value. I checked the spearman correlation between crop area and all x variables, the correlation was good and significant.
My first question is- do I need to consider this equation as time series forecasting or a simple regression problem?
Now, I am planning to fit a random forest regressor. I noticed that when I provide the training data with shuffling = True, it predicts with great accuracy but if shuffling = False, the accuracy is very poor. My second question is- why does this happen? I did not provide time as an explanatory variable, how does the model know? Is it because some of the x variables are also highly correlated with time and the model knows about them?
Do I need to use other machine learning algorithms than a random forest? If yes, which one would work better in this situation?
My objective is to not forecast the area in the future but to do a scenario analysis. For example, I want to know what would be the area trajectory if the interest rate is reduced by 50% or profit is increased by 25%.
EDIT Part of the second question is answered here