Classificação de Dígitos Manuscritos

💻 Sobre o projeto

Classificação de dígitos manuscritos MNIST usando sklearn e Keras.

Dataset

Dataset MNIST de dígitos manuscritos.

Requisitos

• linux system (Ubuntu 20.04)
• python 3.8.10
• matplotlib==3.4.3
• numpy==1.21.4
• pandas==1.3.4
• scikit-learn==1.0.1
• seaborn==0.11.2
• tensorflow==2.7.0

Modelos

Keras 01

Estrutura

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
flatten (Flatten)            (None, 784)               0         
_________________________________________________________________
dense (Dense)                (None, 128)               100480    
_________________________________________________________________
dense_1 (Dense)              (None, 10)                1290      
=================================================================
Total params: 101,770
Trainable params: 101,770
Non-trainable params: 0
_________________________________________________________________
None

Acurácia

97.71%

Matriz de Confusão

Keras CNN 02

Estrutura

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d (Conv2D)              (None, 26, 26, 32)        320       
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 13, 13, 32)        0         
_________________________________________________________________
flatten (Flatten)            (None, 5408)              0         
_________________________________________________________________
dense (Dense)                (None, 100)               540900    
_________________________________________________________________
dense_1 (Dense)              (None, 10)                1010      
=================================================================
Total params: 542,230
Trainable params: 542,230
Non-trainable params: 0
_________________________________________________________________
None

Acurácia

98.62%

Matriz de Confusão

Sklearn KNN

Acurácia

96.88%

Matriz de Confusão

Notebooks

Exemplos

Referências

• Tensorflow: https://www.tensorflow.org/tutorials/keras/classification
• Sklearn: https://scikit-learn.org/stable/