Exploratory Data Analysis

Descriptive Statistics

In this section, we selected category and all audio feature attributes except key to conduct our first round of analysis. Key attribute is dropped since it is highly correlated to mode, an indication of major or minor key, which summarizes Key. The basic statistics of the audio feature attributes offer some insights on the distribution of the variable. Acousticness(slight right skewness), Danceability(slight left skewness), Energy(slight left skewness), Liveness(slight right skewness), Tempo(slight left skewness), Time Signature(slight left skewness) and Valence(slight left skewness), whose means and medians are rather close, have less skewed distribution than InstrumentalnessI(extreme right skewness) and speechiness(right skewness). Meanwhile, since Acousticness, Danceability, Energy, Instrumentalness, Liveness, Speechiness and Valence are normalized by Spotify, the standard deviations reflect the Kurtosis of the distributions of the above variables. Lower(higher) standard deviation implies that more(less) values are around the mean, and thus the distribution will likely to have a thinner(fatter) tail. In addition, the category modes show that ‘party’ and ‘workout’ are the most common categories in our dataset. Other categories all appear in between 2000 to 5000 times.