I have the following time series data:

$\{ t_i, X_i, Y_i \}$

where $i$ is the index, $t_i$ is the timestamp, $X_i$ the measured value of the external variable and $Y_{i}$ the value of the variable of interest. There are several variable components in $Y_i = (y_j)_i, j \in \{1,..,N\}$). At any point in time $i$, only one one of the components of $Y$ was measured. That means, that for example for $i=10$, the data looks like this:

$ t_{10} = $ 2019-04-23 12:34:56

$X_{10} = (2.2, 5, 67, 42, 123)$

$Y_{10} = (nan, .., 4567.89, .., nan) $

I want to train a forecasting model, that predicts the value of $ Y $, even when there is no single time step in the data that contains all the values for every component. So basically $\hat{Y} = (\hat{y}_1, .. , \hat{y}_N)$ with $\hat{y}_j <> NaN, \forall j$. Is there a standard way to deal with such incomplete data measurements?


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.