The dataset is a custom subset of 10,000 songs from the Million Song Dataset, and the labels were obtained from AllMusic.com. Each song has 264 audio features: 48 values for Mel Frequency Cepstral Coefficients(12 bands, 4 statistics), 48 values for the chroma semitones (12 bands, 4 statistics), and 168 values for the rhythm patterns (24 bands, 7 statistics).

The task is to build a classifier that categorize the music into 10 genres based on its audio features.

• Final paper
• Kaggle competition page

Unfortunately my codes are forever gone with damaged computer, and I was too inexperienced to back it up in github.