Decision Tree Induction on the Million Song Dataset -- Modeling Music

Data mining has useful classification methods for the data analysis and prediction. One of them is decision tree induction, which is the learning of decision trees from the class-labeled dataset. It can provide an easy way to understand the data and view the relationship among attributes because it has a flowchart-like tree structure. When I applied the decision tree algorithm with parameters (criterion: gain_ratio and minimal gain: 0.03) to MSD dataset using the RapidMiner tool, the "start_of_fade_out" attribute is the best one to partition the data, as shown in Figure 1. Only 2 Rock and 1 New Age songs have start_of_fade_out that is greater than 547.698 seconds.