Facial expression classification model

This model is designed for facial expression classification and it uses custom CNN model to classify the images into 7 different categories.

This CNN Model is to classify the facial expression into one of the following categories:

Anger
Disgust
Fear
Happiness
Neutral
Sadness
Surprise

Architecture Summary

Layer (type)	Output Shape	Param #
conv2d (Conv2D)	(None, 46, 46, 32)	320
max_pooling2d (MaxPooling2D)	(None, 23, 23, 32)	0
dropout (Dropout)	(None, 23, 23, 32)	0
conv2d_1 (Conv2D)	(None, 21, 21, 64)	18,496
max_pooling2d_1 (MaxPooling2D)	(None, 10, 10, 64)	0
batch_normalization (BatchNormalization)	(None, 10, 10, 64)	256
dropout_1 (Dropout)	(None, 10, 10, 64)	0
conv2d_2 (Conv2D)	(None, 8, 8, 128)	73,856
max_pooling2d_2 (MaxPooling2D)	(None, 4, 4, 128)	0
batch_normalization_1 (BatchNormalization)	(None, 4, 4, 128)	512
dropout_2 (Dropout)	(None, 4, 4, 128)	0
conv2d_3 (Conv2D)	(None, 2, 2, 128)	147,584
flatten (Flatten)	(None, 512)	0
dense (Dense)	(None, 96)	49,248
dropout_3 (Dropout)	(None, 96)	0
dense_1 (Dense)	(None, 96)	9,312
dropout_4 (Dropout)	(None, 96)	0
dense_2 (Dense)	(None, 64)	6,208
dense_3 (Dense)	(None, 7)	455

Total params: 306,247 (1.17 MB)

Trainable params: 305,863 (1.17 MB)

Non-trainable params: 384 (1.50 KB)

Training details

Dataset: https://www.kaggle.com/datasets/manishshah120/facial-expression-recog-image-ver-of-fercdataset

Name	Value
Input shape	48x48 (48, 48, 1)
Optimizer	Adam
Loss	Crossentropy
Max epochs	200
Early stopping monitor	val_loss
Early stopping patience	12

Model performance

Training Accuracy: 0.5758 (Epoch #84)
Training Loss: 1.1272 (Epoch #84)
Validation Accuracy: 0.5823 (Epoch #84)
Validation Loss: 1.1285 (Epoch #84)

Classification report

              precision    recall  f1-score   support

           0       0.52      0.40      0.45       491
           1       0.00      0.00      0.00        55
           2       0.43      0.17      0.25       528
           3       0.83      0.84      0.83       879
           4       0.51      0.67      0.58       626
           5       0.39      0.58      0.47       594
           6       0.73      0.72      0.73       416

    accuracy                           0.58      3589
   macro avg       0.49      0.48      0.47      3589
weighted avg       0.58      0.58      0.57      3589

Notebook

Training notebook: https://www.kaggle.com/code/harkishankhuva/facial-expression-classification