I'm on a team that is tackling a project similar to the following. Suppose you want to estimate the age of a plant using a small set of tabular data features. In addition to the plant data, you have weather data for all points in time, which is highly relevant to the plant's age. (Two plants with similar features may have very different ages based on their weather histories.)

Are there any well-established methods for incorporating the weather data in a case like this? We obviously can't include all our weather data as regressors for each plant -- we have decades of weather data, and while some plants might be several years old, others might be only days old.

The best I've been able to come up with so far is to engineer a small set of features for each plant that will be used as regressors. These will be a few values summarizing the weather for each of five different time scales: the previous day, the previous week, the previous month, the previous year, and the previous five years. This seems like an unhappy solution, though, as it is both more info than will be relevant to very young plants, and much less info than we have access to that is relevant to older plants.

The only other idea I've had is to build a recurrent NN that "de-ages" the plant one day at a time until the plant reaches estimated age 0. Any tips would be appreciated, I find this a very difficult problem even to research, because I can't think of good search terms that narrowly define this issue!

  • 1
    $\begingroup$ What you describe is fine for creating the weather-related variables (aggregating over different time periods). I guess the only caveat is that the weather-related variables will be highly correlated with the response, which is the age of the plant. I'd therefore expect these variables to dominate the regression/model you estimate. Once you've created these variables, I'd start simple (standard linear regression). Assess the performance of these simple models against more complex ones. Iterate until you're satisfied. $\endgroup$
    – ralph
    Jan 8 at 3:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.