Title: Make My Playlist

Group Members: Seher Khan, Sudeeksha Sanikommu and Shreyasi Chaudhary

Goal: The aim of this project is to make it easier for a user to generate their playlist given a set of songs.

We employed a classification approach which creates the playlist based on predefined playlists and obtained satisfactory results. We also experimented with a clustering approach, taking into account the listening history of similar users and producing more personalized playlists.

Approach 1: Number of rows (songs): 8980 Features: popularity, acousticness, energy, liveness, tempo, speechiness, instrumentalness, time_signature, danceability, key, duration_ms, loudness, valence, mode Classes: Love, romantic, classical, edm, dance, breakup, rock, jazz, sad, trance, hiphop, country, party, new hits, workout, old hits, happy Source: Spotify

Classification models employed: Decision Tree, SVM, Logistic Regression, Naive Bayes, LDA, QDA, Random Forest, Gradient Boosting Classifier, One vs Rest Classification, Neural Networks, Ada boost, Voting Classifier, Class Hierarchy

Approach 2: Data:

• Listening history of 54,687 users was obtained from the Echo Nest.
• Users with fewer than 5 songs in their history were dropped.
• Features of each song was extracted for each remaining user and then averaged * out to produce features for the user.

Clustering model employed to cluster users: Mean Shift Clustering, K means, and Affinity propagation