A data scientist computes overall model accuracy without accounting for class imbalance. Which concept explains why this metric can be misleading?