Do random forest, K-means, SVM take into consideration past value in time series ?

Dear all, sorry for my stupid question but I am new to machine learning.

I was wondering if I should introduce lagged variables in my series to take into consideration past information.

If it helps, I am doing a classification on stock performance forecasting (either negative, neutral or positive). Therefore, each line correspond to a month with its different observations (predictors).

After normalising them, I don't know if these algorithms take into consideration past values, in other words if they recognise that some indicators are particularly high of low, compared to previous months and take a wise decision in function of that.

I had a doubt since for a tree, decisions points are made with the "best" threshold (gini). Did it then took into consideration all past values ?

Many thanks in advance,

Pierre

Answer 1

[Bernhard Suhm]

Your model takes into consideration whatever you provide as predictor variables with your data. None of the machine learning methods you mention "automatically" incorporate past data points from your time series in making predictions, you have to "feed" them into your algorithm, as "lagged variables" as you say.

[Pierre Lonfat]

Thank you very much for your answer. I thought that decision trees would use past values to calculate the best threshold when making decision at each node (the gini index). Is that incorrect ? I am going to lag my variables anyway :) !