ProblemsML0005

Sampling Bias in Event Collection

Easy

Problem

A model trained on production events keeps underperforming for exactly the users it needs to work for. The culprit isn't the model; a post-mortem traces it to collection: whichever of the 8 traffic segments generates the most raw events dominates the training set, at up to 2,000 events/second during peak hours, while rarer segments are barely represented at all. Fix it at collection time: deliberately balance what gets kept across segments, and make the resulting policy visible enough that someone can actually audit it.

Required components

Click to add a component to the canvas.

Additional components

Browse by category. Click or drag a component onto the canvas.

Drag components from the library onto the canvas, then drag from a component's right handle to connect it to another.

100%

Design canvas

Sign up to start designing

Sign up free

Premium is free for 58 more days.