Group Members-
- Achintya Gupta
- Ankita Kanoji
- Arnav Sharan
- Kannan Rustagi
More details of the ideas, implementation and results can be found in the final report.
We have devised an innovative approach to predict a track’s mood based on audio features. We have mainly classified a song into one of four moods – calm, happy, energetic, and sad. For EDA, we have constructed Heatmaps correlating the features, Radar plots, and KDE plots for better visualization of our data. We attempted to do feature selection using Principal Component Analysis. We have implemented Classification as well as Clustering methods from scratch in python to model our data. The Classification models include the Gaussian Naive Bayes classifier, Decision Tree, Random Forest Classifier, and K-Nearest Neighbours. We have contrasted the results we obtained from our custom implementation with those obtained by using Sklearn's implementation as well. As for clustering, we have employed the K-Means algorithm. After this, we compared the results obtained from both paradigms and attempted to draw conclusions on our dataset