Can audio predict a hit?
Artists and producers decide how loud, long, and fast a track should be. We wanted to know whether those choices tell them much about how a song might perform on Spotify.
We modeled Spotify's 0–100 popularity score using 32,833 track records. We tested each feature separately, compared models with and without tempo, and checked how far individual scores could fall from the fitted trend.
Why the forecast falls short
Even with all 3 features, the model's errors had a spread of about 25 points on a 100-point scale. That makes the results useful for describing the sample, but too weak to set production targets or forecast a release.
Loudness
Louder tracks tended to score slightly higher. But the scatterplot shows why that average trend isn't a recipe: tracks at similar loudness levels scored very differently.
Tempo
Adding tempo barely changed model fit: the share of variation explained rose from 2.236% to 2.245%. The model comparison reported p = 0.081, which does not meet a 5% significance threshold.
Duration
Track length was associated with popularity after accounting for loudness and tempo, but length alone still left a wide range of plausible scores. The analysis gives no basis for prescribing an ideal song length.
