Machine Learning Model on Spotify

Song Popularity

The machine learning model predicts song popularity amongst listeners on Spotify.  A subset of 12,028 observations from a larger dataset with 18,835 songs are used to train an Artificial Neural Network algorithm. Quantitative factors that influence song rating are key, tempoenergy, danceability, audio mode, duration, liveliness, loudness, time signature, vocal prominence, speech frequency and acoustic confidence measure. The predictive model usefully quantifies song characteristics that likely determine overall streaming volume on Spotify over an extended period of time. As underlying factors that are positively associated with song popularity are analysed, an up-and-coming songwriter would be able to apply insights from the predictive model to optimally allocate music production resources while songwriting for a desired target audience. 

Most Influential Song Characteristics

The three production elements that likely garner frequent and repetitive streams are average loudness over the duration of a track, the perceptual level of energy the song amplifies and the vocal consistency of an artist relative to instrumental interludes. Energy and loudness are positively correlated. As the intensity of a song mounts sound amplitude oscillations equally heighten. A song with an average sound pressure level of -15 decibels and lower may aptly engage an unassuming listener with a keen ear for inherent song volume.  Dynamic and upbeat tracks with consistent vocal idiosyncrasies boast above average listenership. Contrarily, low fidelity or acoustic songs with minimal or absent vocals corner a niche set of listeners. Songwriter resources would benefit from prioritising energetic and robust sound elements including overall tempo regularity and beat strength, coupled with stable and rhythmic vocals to widen audience reach.

Songwriter Popularity Boost

The time signature, key and audio mode of a song all have a similar effect on streaming popularity therefore a new artist has sufficient leeway to trial and test beats per bar, song pitch and modality whilst composing resonant melodies and song lyrics. The cumulative consideration of the aforementioned song characteristics ought to safeguard a popularity score that approximately ranges between 40 to 67. Song length is likely the eleventh most significant characteristic to consider as the duration of a track seldom has a dire impact on listener interest. However, songs that are roughly between 211666 - 218304 milliseconds may be a helpful time reference.

Correlative Relationship Between Music Production Components

Distribution of Song Popularity