I am working on a dataset contains one output variable and a number of input variables.The data looks like the following:
Y, X1, X2, X3, X4
7, 5, 0.7, 8, 9
3, 6, 0.3, 9, 9
....
Where Y is the output and X1 to X4 are the inputs.
The order of the data is important. I am assuming for example that the Y value (3) is affected by the current X values (6 0.3 9 9) and previous X values (8 0.7 8 9). Then, I trained a Random forest model based on this idea by considering the current X values and previous X values as input variables. The testing and traning MSE error of the model is better compared to a model without considering the pervious inputs effects.
My question that, is my methodology valid? Should I consider other algorithms such as LSTM or RNN? The problem of RNN and LSTM in my case is that my dataset is not a series data. The output is only affected by the nearest previous output and current output. What are your thoughts?
