A project accepted in APOGEE (BITS Pilani’s Technical Fest). Work has been done to create a machine learning model in order to correctly predict Latin music and to reduce the subjectivity of music genre classification that currently exists in the music industry. Some features that we selected were mel-frequency cepstrum coefficients, filterbank energies and spectral subband centroids. These features were then broken into spatio-temporal regions. After doing so, SVM classifiers were trained for each temporal segment of the 1000 songs. In this process we make a prediction matrix for each song - classifying it's genre by analyzing the spatial features of a particular temporal slab. Finally this prediction matrix is fed to a random forest classifer resulting in 70% accuracy (7 times the base line).