1

I am new to time series forecasting and all the articles that I read online talk about time series forecasting using a variable that is tracked over time and then forecasted after the model is trained. These examples include - sensor readings over time, daily stock price movements, temperature swings etc.

What I am looking to learn is - how can we include other predictors/independent variables in time series forecasting? Comparing it with linear regression or supervised problem framing in general, for instance, we can have one hot encoded variables as predictors in time-series model.fit() ? How does this thing work?

Since I am new to this time series forecasting domain, I would appreciate if someone can point me to right resources/links/blogs. Thanks in advance.

Here is the problem that I am facing in the marketing domain - given some historic data about how marketing leads are generated, I want to forecast how many leads would be created in next 2/3 quarters. So this is not a traditional linear regression problem where I have X variables and I try to predict my target variable y which is #LeadsGenerated in this case. Neither this is a straight forward Time Series forecasting problem where I have some temporal features that I can fit a model upon.

What business needs to understand is - given the historical data (like marketing spend data, campaign performance data, news articles etc) for last 3 years, how many leads will be generated in future?

Regressor
  • 121
  • I'm not sure if this applies, but you can often restate the problem in other terms. For example, the SIRD epidemiology model is non-linear when expressed as time derivatives, but becomes linear if appropriate substitutions are made to change the independent variable. Track the number of deaths instead of time to simplify the problem. – TurlocTheRed Feb 17 '22 at 05:36
  • thank you for the advice. What I am facing at the moment is to forecast how many marketing qualified leads will be generated in next 2/3 quarters. For this I want to use a forecasting approach but my independent variables are a mix of numeric and categorical variables too. So I am not entirely sure on how to use them in time series. If you happen to have any links, I would really appreciate it. I have added my business problem too , just to make things more clearer. – Regressor Feb 17 '22 at 05:42

0 Answers0