A binary classifier answers a yes-or-no question. The input starts small (8 features per sample), and the network expands it through hidden layers: 8 → 16 → 32 → 64, before collapsing to a single output.
That last dimension, 1, is the prediction. After applying a sigmoid function, it becomes a probability between 0 and 1. We can pick a threshold (typically 0.5) to convert the probability into a binary prediction. Above 0.5, the answer is yes. Below 0.5, no.
Watch how I have drawn the matrices getting taller as we go deeper, then collapsing to a single row at the output. The network first builds a richer representation, then makes a decision.
Next:
5. Encoder