Beating a Baseline With a Fixed LinearSVC Using Only Data-Side Changes

MediumClassical ML

Reported in interviews at DRW

Problem statement

How would you beat a classification baseline while keeping LinearSVC as the final estimator? The table mixes numeric fields on very different scales, categorical values and short text. Duplicate rows sometimes disagree on labels, and all model choices must use training data because hidden test feedback stops after you beat the baseline.

Improve preprocessing and validation, then write reproducible training and prediction functions.

Show hint

Since the estimator cannot change, ask what a linear large-margin model needs from each kind of column to make good use of it.

Relevant roles: ML Engineer, Data Scientist and Applied Scientist.

Related free problems