MNIST-Digit-Recognition

The popular MNIST dataset is used for the digit recognition task using different machine learning algorithms such as KNN and SVM with HOG features. A simple feed forward neural network is also used for comparision with the machine learning models. A detailed report in IEEE format is also provided in the documents folder.

Results

Model	Accuracy
KNN (k=3)	99.97
SVM (poly kernel)	98.38
FFNN	97.98

Directory Structure

📦MNIST-Digit-Recognition
┣ 📂Documents
┃ ┗ 📜Detailed Report.pdf
┣ 📂Notebooks
┃ ┣ 📜MNIST using FFNN.ipynb
┃ ┣ 📜MNIST Using KNN.ipynb
┃ ┗ 📜MNIST Using SVM and HOG.ipynb
┣ 📂Trained models
┃ ┣ 📜linearSVM.pkl
┃ ┣ 📜nonlinear-Poly.pkl
┃ ┗ 📜nonlinear-Rbf.pkl
┗ 📜README.md
 

To Do

• Use Convolutional Neural Networks (CNN).
• Read digits from images using OpenCV.