There comes a point in every data professionals career where the usual models start feeling a little too simple. You run a linear regression, the results look fine, but something about the prediction...
Updated 1 month ago
Version 1.0
Blog Post
This is such an important mindset shift Sahir!
Most analysts get stuck treating spikes as "noise" or "outliers". Reframing them as event-driven signals changes everything. The point about making performance metrics more "fair" by accounting for triggers really hit home.
Poisson regression is perfect here because it models counts and rates, not averages. I’ve used the same approach for support ticket volume - once we added "product release" and "system incident" as features, the model went from 62% to 89% accuracy.
The boxplot + violin plot visuals you shared also help stakeholders *see* the spread and bimodality that averages hide.
Thanks for sharing this. Following you for more such practical data science insights!