I currently have data from several studies, with each having different sample sizes and possibly different set-ups. There is a common binary variable of interest across all studies that I would like to predict. Suppose I have $5$ studies, $s \in \{1, \ldots, 5\}$, with each study having a $Y_s$ vector of binary outcomes, and an $X_s$ matrix of predictors in the columns and observations by row. One idea I have is to train a classifier method on each of the data pairs across the $5$ studies:

$$ (Y_1, X_1), \ldots, (Y_5, X_5) $$

Then somehow aggregate them. What becomes complicated is how I should think about the training/testing set breakdowns. In other words, should I break the sets according to study and if so, how can I combine them later?

I am wondering generally if such frameworks exist for machine learning methods across many studies. Thanks.


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.