3
$\begingroup$

I have a set of predictor variables and another target variable .

Now I am really confused on what method to use to forecast the target variable .

For e.g my data set have customer profit(which is my target variable) and a set of predictor variable(balances of different account) for one year for each customer .

Now I need to predict profit of next 5 years .I am confused in the part that I dont have the data(predictor variables) for future .

What are my possible choices of modelling .Please assist .

Improve this question
$\endgroup$

3 Answers 3

Reset to default
3
$\begingroup$

You should distinct between a time series prediction, where from a known history of some attribute the future is predicted and model prediction where based on the predictor variables the target variable is calculated.

In your case you could combine both approaches, i.e. use time series prediction on the customer balances and apply the regression model to calculate the profit on the result.

Improve this answer
$\endgroup$
2
  • $\begingroup$ I liked this idea . currently building the model . $\endgroup$
    – Bg1850
    Commented Dec 9, 2015 at 20:51
  • $\begingroup$ The problem here is how to incorporate the uncertainty on your predicted covariates into the regression model, which will have its own uncertainty... $\endgroup$
    – Spacedman
    Commented Dec 23, 2016 at 18:27
0
$\begingroup$

One way to generate a set of predictor variables is by adding noise. In your case, this might work well unless there is a lot of variation due to uncontrolled conditions like financial crisis. You have to be careful about adding noise. It should thoroughly test your model for robustness. One way would be to add fractions of the variance of each attribute.

Improve this answer
$\endgroup$
2
  • $\begingroup$ I already have a set of predictor variable . I dont understand why I would add noise in this scenario . $\endgroup$
    – Bg1850
    Commented Dec 4, 2015 at 17:08
  • $\begingroup$ @Rishiraj, I have seen this idea of adding noise thrown around a lot when it comes to unreliable variables. It sounds interesting, but do you have any evidence that it works? Papers? $\endgroup$
    – Ricardo Cruz
    Commented Dec 23, 2016 at 14:15
0
$\begingroup$

For simple extrapolation, use the predict() function; call it with the newdata= Argument. This is a dataframe with the attribute values you for the next n years.

Otherwise, check the forecast() function from the forecast package. It is recommended by the R cheatsheet here

Improve this answer
$\endgroup$
3
  • $\begingroup$ I dont have information of next year's data .No way predict can give any prediction without having a data for the predictor variable .Forecast uses a linear model (ARIMA is linear) so it gives very crude predictions . However that is going to be my first step towards solving the problem . a time series model for the predictor's then a predictive model for the target $\endgroup$
    – Bg1850
    Commented Dec 9, 2015 at 20:50
  • $\begingroup$ you can pick some reasonable values for the attributes for the future (or assume last years' attributes remain constant) . good luck. $\endgroup$
    – knb
    Commented Dec 10, 2015 at 8:18
  • $\begingroup$ can you quantify "reasonable " ? $\endgroup$
    – Bg1850
    Commented Dec 10, 2015 at 18:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.