Update classifiers.py

[mst-rajatmishra]

Hi, I tried making some changes,

• The optimizer now correctly uses learning_rate instead of lr.
• BatchNormalization layers have been moved before activations.
• The loss function has been changed from mean_squared_error to binary_crossentropy, which is the appropriate choice for binary classification with sigmoid output.
• The model summary is printed during initialization for debugging.
• The dropout rate is now a parameter of the constructor for all models (dropout_rate), allowing for easy tuning.
• Replaced test_on_batch with evaluate, which is more suitable for evaluating the entire dataset.
• Added HeNormal initializer for weight initialization in convolutional layers. Thank You, RAJAT MISHRA.