How can I use a multivariable list of strings as a single X variable / feature in a Neural Network Model

Question

Here is an example of what I am trying to do.

I am using a bunch of X variables to predict a Y.

X1 = Number of days since applying first medication X1.1 = Total times the medication should have been applied (calculated using frequency in prescription) X1.2 = Total times the medication was applied (calculated)

X2 = Number of days since applying first medication X2.1 = Total times the medication should have been applied (calculated using frequency in prescription) X2.2 = Total times the medication was applied (calculated)

X3 = Number of days since applying first medication X3.1 = Total times the medication should have been applied (calculated using frequency in prescription) X3.2 = Total times the medication was applied (calculated)

Y = Skin clear 0 / 1

I have some sample data as below:

Now, the X1.1 and 1.2 are descriptive of X1. As a separate input to the Neural Network, they are meaningless.

I am trying to find, how I can tell the model that X1, X1.1 and X1.2 together is a single input with multiple dimensions.

I use the below code:

classifier = Sequential()

classifier.add(Dense(units = 14, kernel_initializer = 'uniform', activation = 'relu', input_dim = 27))

classifier.add(Dense(units = 14, kernel_initializer = 'uniform', activation = 'relu'))

classifier.add(Dense(units = 1, kernel_initializer = 'uniform', activation = 'sigmoid'))

classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

classifier.fit(X_train, y_train, batch_size = 25, epochs = 500)

Any help will be really appreciated.

Answer 1

For your problem you can easily use PCA to remove the redundant features and only keep the viable information. Since you are using neural networks, you anyways wont have any interpretation layer, so PCA will not be a problem.