Final Project Machine Learning Course

By: @kggold4 @TalSomech

Spotify dataset classification

Dataset link

Mark: 96

Code

Our main project is in the spotify_classification.ipynb notebook, also see the utils.py file (for our utils functions).

Goal

The goal of this project is to train machine learning models (supervised) that will classified the popularity of a spotify song to three classes:

1. high popular
2. medium popular
3. non popular

Dataset:

Features:

1. acousticness (Ranges from 0 to 1)
2. artists (List of artists mentioned)
3. danceability (Ranges from 0 to 1)
4. duration_ms (Integer typically ranging from 200k to 300k)
5. energy (Ranges from 0 to 1)
6. explicit (0 = No explicit content, 1 = Explicit content) - Categorical.
7. id (Id of track generated by Spotify) - Numerical.
8. id_artists.
9. instrumentalness (Ranges from 0 to 1).
10. key (All keys on octave encoded as values ranging from 0 to 11, starting on C as 0, C# as 1 and so on…).
11. liveness (Ranges from 0 to 1).
12. loudness (Float typically ranging from -60 to 0).
13. mode (0 = Minor, 1 = Major).
14. name (Name of the song).
15. popularity (Ranges from 0 to 100).
16. release_date (Date of release mostly in yyyy-mm-dd format, however precision of date may vary).
17. speechiness (Ranges from 0 to 1).
18. tempo (Float typically ranging from 50 to 150).
19. time_signature.
20. valence (Ranges from 0 to 1).

Correlations between popularity and others features

NOTE: during the ordering of the data we applay the popularity for classification to be in the following format:

class	real value	class value
high popular	70 <= x	2
medium popular	40 <= x < 70	1
non popular	x < 40	0

In order to see the distribution between the number of popularitry classes (unbalanced number of features in data):

Models result (unbalanced number of features in data):

model	accuracy
KNeighbors Classifier	74.20 %
Logistic Regression	72.32 %
XGB Classifier	77.74 %
MLP Classifier	70.82 %

Models result (balanced number of features in data):

model	accuracy
KNeighbors Classifier	59.35 %
Logistic Regression	60.06 %
XGB Classifier	65.41 %
MLP Classifier	64.16 %

Models result between regular and balanced data trained (shown in the notebook):

Conclusions:

It's very difficult to precdict popularity of spotify tracks with the data we have in our data set, Even after we cleaned & normalized our data, and creation balanced and non-balanced training data for our models, We still see that the accuracy of our models is moderate.