Question

A data scientist is preparing a dataset to predict whether a customer will default on a loan within the next 90 days. The raw table contains a column named 'days_until_charge_off', which records how many days after the loan the account was written off as a loss (populated only for accounts that eventually defaulted). While building the training pipeline, the scientist notices the model achieves near-perfect validation accuracy. What is the most likely problem, and what should be done?

Accepted Answer

The 'days_until_charge_off' column causes target leakage because it is only known after the outcome occurs; it should be dropped before training.

Answer

The validation set is too small; increase its size to reduce the inflated accuracy.

Answer

The features need to be scaled with StandardScaler to prevent the model from overfitting.

Answer

The random seed was not set, so the near-perfect accuracy is a coincidence that will disappear on the next run.