The project implementation for EEE485 course, Bilkent University.
- @emredonmez98
- Used Spotify Tracks DB dataset.
- Removed severely underrepresented classes, and shaved the samples sizes to make the equal in size and normalized data in preprocessing.
- Used different features of the data since we have used a dataset with both categorical and numerical features.
- The discussed techniques are applied from scratch, without using any ML/DL framework.
- Implemented in Python (Jupyter Notebook)
- Used techniques are
- kNN
- Random Forest and
- Multilayered Perceptron (MLP) / Feed-Forward Neural Network
- Varying Number of layers and neurons
- SGD, Momentum, Adam and AMSGrad Optimizers
- Glorot and He Initializations
- Cross Entropy and Mean Squared Error Losses
- Sigmoid and ReLU activations
- Repeated the experiments using PCA features and compared.