You're building a binary classification model where the training data is highly imbalanced: 99.8% of samples belong to class 0 (negative), and 0.2% belong to class 1 (positive). To address this imbalance, you down-sample the majority class (class 0) by randomly selecting 1% of its samples, while retaining all samples from the minority class (class 1).

After training the model on this balanced subset, you now need to predict probabilities for the original imbalanced population. How would you adjust the model's output probabilities to ensure they reflect the true likelihood of class 1 in the real-world population?