Logistic Regression - Decision threshold

• The classification threshold (Decision Threshold) isn’t always 0.5 in all the systems always. (Btw default is 0.5)
• A cutoff value … <refer slides..>

When to Lower the Threshold (e.g., to 0.3)

• We should lower the threshold when the cost of a False Negative is very high. A False Negative is when we predict “No,” but the real answer is “Yes.”
• False Negative - means predict values as negative (-1) btw their actual value is positive (+1).
• If False negative is very high, then it need to reduce the decision threshold (less than 0.5).
• Goal: - - -

When to Raise the Threshold (e.g., to 0.8)

• We should raise the threshold when the cost of a False Positive is very high. A Flase Positive is when we predict “Yes,” but “No.”
• Goal: - - -
• False Positive - means predict values as positive (+1) btw their actual value is negative (-1).
• If False positive is very high, then it need to increase the decision threshold (more than 0.5).

Confusion Matrix