Music recommendation comes under category of Music Information Retrieval (MIR) which has been quite a topic of interest these days. Music is categorized by various features including rhythmic structures, member form and instrumentation. To determine the interest of the user is a big challenge for the MIR community. Through the means of this research paper we aim to present a music recommendation system, which provides personalized recommendations to each user. It is based on users past likes and listening history. The features are first extracted from the database of audio files which are in .au format. The proposed model is used to train these audio files and different clusters are formed accordingly allotting the songs in the database to predicted category. Now the user liked songs features are extracted and the built model predicts the recommendations for the user by matching the category allotted to this song, to that of other songs. An accuracy of close to 75% was achieved during the course of the project.