The aim of this project is to implement different conventional classifiers to achieve hand-written digits recognition, as well as performing transfer learning in an image classification task.
Following Algorithms were implemented:
- Support Vector Machines (dimensionality reduction with PCA and MDA)
- Multiclass Logistic Regression (dimensionality reduction with PCA and MDA)
- Convolutional Neural Network
Transfer Learning has been implemented on the monkey species dataset.
- Support Vector machines:
- Linear Kernel - 93.68% (combined with PCA)
- Polynomial Kernel - 97.19% (combined with PCA)
- Radial Basis Kernel - 96.55% (combined with PCA)
- Multiclass Logistic Regression - 92.56% (combined with PCA)
- Convolutional Neural Network - 98.84%
- Transfer Learning - The accuracy of a new model was increased from 51.66% to 95.22% when replaced with MobileNet model with pretrained weights on imagenet dataset.
- Tensorflow
- Scikit-Learn
- Seaborn
- Numpy
- Matplotlib