- Engineered an XGBoost Regression model to predict the popularity of a song on a scale of 1-100. The model was able to predict popularity with a mean absolute error (MAE) of 9 points and could help music producers and artists in foreseeing popularity of their songs. Highlights -
- Implemented K-means algorithm to cluster songs to see if songs with similar popularity would cluster together. They didn't.
- Tried several algorithms such as linear regression, KNN regression, XGBoost regression, and Neural network and identified XGBoost as the best model, based on its superior performance on the validation dataset.
- Additionally, tried a multi-class classification technique, using XGBoost classification, to categorize the popularity of a song, instead of predicting the exact value.
- Designed a Spotify’s song radio like feature to recommend top 5 similar songs, using Principal Component Analysis (PCA) and the Nearest Neighbors algorithm.