Class Imbalance Quant Interview Guide
Class imbalance quant interview guide for rare events, metrics, thresholds, sampling, costs, calibration, and examples.
Candidates working with rare events, binary labels, or skewed outcomes.
Imbalance changes metric meaning
When one class is rare, accuracy can look strong while the model misses the important cases. Metrics must match the decision and class costs.
Use threshold-aware thinking
Precision, recall, false positive costs, false negative costs, and calibration often matter more than a single default threshold.
Concrete example
If only one percent of events are positive, a model that always predicts negative is 99 percent accurate and still useless for finding positives.
Sampling affects probabilities
Oversampling or undersampling can help training, but predicted probabilities may need recalibration before being used for decisions.
Common mistakes
Candidates often optimize the metric that looks best. Explain which errors are costly and how you would choose a threshold under those costs.
Practice the pattern
Use the LeetQuidity curriculum and calibration to turn this topic into a focused practice plan.