shahabty / hit_song_prediction

Hit Song Prediction

This repo contains implementations of random forest, support vector machine, and neural network for hit song prediction in Pytorch and Sklearn.

Installation

1. Install Anaconda3.

2. Run the following commands to create conda environment and install all dependencies:

username@PC:~$ conda env create -f environment.yml
username@PC:~$ conda activate hits

Training and Testing

Neural Network Model

username@PC:~$ python train_nn.py.

SVM Model

username@PC:~$ python train_svm.py

Random Forest Model

username@PC:~$ python train_random_forest.py

Visualization

Please visit this link for the visualization of our neural network results.

The code for the above visualization is available here.

Evaluation

Our evaluation metrics are precision, recall, and F1-Score:

Model	Precision	Recall	F1-Score
SVM(High)	0.71	0.70	0.71
SVM(Low)	0.71	0.80	0.75
SVM(Combined)	0.71	0.80	0.75
RF(High)	0.90	0.93	0.92
RF(Low)	0.92	0.94	0.93
RF(Combined)	0.91	0.94	0.93
NN(High)	0.87	0.82	0.84
NN(Low)	0.90	0.87	0.89
NN(Combined)	0.90	0.90	0.90

High: only high-level features are considered in the training and testing. Low: only low-level features are considered in the training and testing. Combined: Both high-level and low-level features contribute to the prediction.

Contributors

Mahshid Alinoori, Sara Boback.