This is the repository for the final project for MSCA Machine Learning and Predictive Analytics Course. In this project, we tried to classify the 10 classes of the fashion dataset using various traditional machine learning, deep learning and transfer learning models.
This project requires Python 3.7 or above and the following libraries installed -
- NumPy 1.16.5
- Pandas 1.3.1
- matplotlib 3.1.1
- scikit-learn 0.22
- seaborn
- opencv
- xgboost
The transfer learning models required GPU with large memory.
All our finalized notebooks are in the "notebooks" folder.
-
Notebook mnist_fashion_analysis.ipynb -
- This notebook contains the EDA, Machine Learning and Deep Learning Models.
- Linear regression, SVM with RBF Kernel, Gradient Boosting, XGBoosting, Simple NN, CNN with one Conv2D, CNN with two Conv2D were implemented.
- The html version contains the outputs from our final models.
-
The following notebooks contain the correponding transfer learning models -
- VGG - VGG16.ipynb, VGG19.ipynb
- InceptionResNetV2 - InceptionResNetV2_0530_V2.ipynb, InceptionResNetV2_0530_V3.ipynb
- ResNet50 - resnet50_model_vh.ipynb, resnet50_v3_model_vh.ipynb, resnet50_v4_model_vh.ipynb, resnet50_v5_model_vh.ipynb, resnet50_v6_model_vh.ipynb
- InceptionV3 - InceptionV3_0530.ipynb
The results folder contains images of the accuracy and loss plots from our models above as well as an image showing the layer used for the CNN with two Conv2D model that had the highest performance with an overall accuracy of 92.6%.