Video Transcript
Those in the music industry spend countless hours trying to crack the code to find a song that will catch the attention of audiences. To ease the burden, researchers from Claremont Graduate University offered a scientific approach that introduced machine learning to predict which songs would be hits and which would be flops.
In their research article, Sean Merritt, Kevin Gaffuri, and Paul Zak showed that applying machine learning to neural data collected from people listening to new music could predict hit songs. The study featured 33 participants that were fitted with cardiac sensors. In groups of five to eight, the participants listened to 24 songs picked by a streaming service and were asked about their preferences.
The selected songs had not been out for more than six months, and the researchers let the streaming service define a hit song as one with over 700,000 streaming listens. The researchers used a commercial platform from Immersion Neuroscience to measure parties’ neurophysiologic reactions to the songs. According to Zak, the collected signals reflected a brain network’s activity associated with energy and mood levels and allowed the researchers to predict market results.
After the researchers collected data, they used statistical approaches to evaluate the predictive precision of neurophysiological variables. They then created a machine learning model that tested various algorithms to reach the highest prediction outcomes.
The linear statistical model found hit songs 69% of the time. But when machine learning was added to the data, the success rate increased to 97%. Machine learning added to the neural responses to the first minute of songs also posted an 82% success rate. The researchers said the approach could be used for more than hit song identification, including television shows and movies.
