PCA proyection for Convnet 1D results

summary(model)

___________________________________________________________________________________________________
Layer (type)                                Output Shape                            Param #        
===================================================================================================
input_1 (InputLayer)                        (None, 45)                              0              
___________________________________________________________________________________________________
embedding_1 (Embedding)                     (None, 45, 100)                         4000           
___________________________________________________________________________________________________
conv1d_1 (Conv1D)                           (None, 42, 256)                         102656         
___________________________________________________________________________________________________
flatten_1 (Flatten)                         (None, 10752)                           0              
___________________________________________________________________________________________________
dense_1 (Dense)                             (None, 1024)                            11011072       
___________________________________________________________________________________________________
dense_2 (Dense)                             (None, 1)                               1025           
===================================================================================================
Total params: 11,118,753
Trainable params: 11,118,753
Non-trainable params: 0
___________________________________________________________________________________________________

Using a PCA 2D projection is possible to observe the FP points for conficker form a cluster along the PC3-axis for the values (1,1) for the PC1 and PC2 axis. Notice that all the Conficker False positives have a character length around 4.