I am building a Recurrent Neural network (LSTM) for predicting the number of days until a Pollen season starts (when the cumulative of the year exceeds X). One of the features I am including in my model is the weather forecast.
However, I do not feel confident about the way I defined the model while including this weather forecast; currently, the weather forecast of 7 days is included as one of the predictors, However, when the label (number of days until the season starts) is smaller than the forecast I am training te model on forecast data which is completely irrelevant in determining the start of the season (e.g. if the season starts in 2 days and I am including the forecast of 7 days as predictor I am also training the model on the 5 days after the season already started while these are completely irrelevant).
My feeling is that it is not right when training RNN's for survival analyses. Does anyone know a way to deal with this? Or have an example where someone dealt with a similar issue?
Thanks a lot!