I have binary classification problem, we know that the output layer will be scalar or dense a.k.a. 1 unit neuron with sigmoid as function activation. 1 means is the subject will die, while 0 means keep alive.
My data type is tabular, where rows represent time (hours) and cols represent features such as gender, age, white blood count, heart rate, and so on. The time is bounded with first 24 hours when collecting data while the feature is limited to the 19 features
Therefore the model was trained with fixed input shape matrix (24, 19) where 24 represent hours and 19 represent features.
While model in inference mode (forward propagation) after trained, can I change input shape dynamically such as (24,10) in case I didn't get some features for all 24 hours such as heart rate. Or, (10,19) in case didn't get all captured data for 24 hours. Or, (1,1) just a hour and a feature such as only heart rate at that time while another are NULL.
Or, shall I fill the NULL value with an interpolation or an average value to achieve fixed shape?
Theoritically, the weights are saved in hidden layers after training, therefore reloading weights after modify input shape is possible. But I have no idea which one shall I chose