[Python] Why is my CNN model overfitting even with dropout and data augmentation?

I'm working on a CNN-based image classification task using Keras and TensorFlow. My dataset contains around 5,000 labeled images across 5 classes. I'm applying data augmentation and using dropout layers, but the model still overfits — training accuracy reaches ~95%, while validation accuracy stays around 70%.

model = tf.keras.models.Sequential([
tf.keras.layers.Conv2D(32, (3,3), activation='relu', input_shape=(128, 128, 3)),
tf.keras.layers.MaxPooling2D(2,2),
tf.keras.layers.Dropout(0.2),

tf.keras.layers.Conv2D(64, (3,3), activation='relu'),
tf.keras.layers.MaxPooling2D(2,2),
tf.keras.layers.Dropout(0.3),

tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dropout(0.5),
tf.keras.layers.Dense(5, activation='softmax')
])

model.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=['accuracy'])
I’m using ImageDataGenerator for augmentation:

datagen = ImageDataGenerator(
rescale=1./255,
rotation_range=20,
zoom_range=0.2,
width_shift_range=0.1,
height_shift_range=0.1,
horizontal_flip=True,
validation_split=0.2
)
What I’ve tried:

Increased dropout rates

Reduced model complexity

Added L2 regularization

Tried early stopping

Played with learning rate

Questions:

What could be the reason for continued overfitting?

Are there other techniques I can try to regularize the model better?

Could this be due to an issue with my dataset or preprocessing?

Any advice or suggestions are welcome!

Continue reading...