am doing my research using lstm-rnn algorithm. i have time-series and non time-series features. how to apply lstm on my dataset? and also how to apply feature selection mechanism to select features?
1 Answer
$\begingroup$
$\endgroup$
Considering that Neural Networks (LSTM-RNN in this case), like the rest of deep learning methods, are like black boxes.
The transformation applied are too complex to determine how much a value, variable, weight affects the solution.
Although is a hot topic, there's no current method viable or wide-known used to understand which variables are better or have higher influence in the model.
However, what you can do is an EDA (Exploratory Data Analysis) between your explanatory variables and response variables. Their relationships can give you some skyview of which variables are more correlated or influenced by others.
