Threshold tuning in ML Python - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding the effect of threshold on classification

In a binary classification model, what happens to the precision and recall when you increase the decision threshold from 0.5 to 0.8?

A Both precision and recall increase

B Precision decreases and recall increases

C Precision increases and recall decreases

D Both precision and recall decrease

💡 Hint

Think about how raising the threshold affects which predictions are labeled positive.

Increasing the threshold means the model is more strict to label positives, so fewer false positives occur, increasing precision. But some true positives are missed, lowering recall.

❓ Predict Output

intermediate

Output of threshold tuning code snippet

What is the output of the following Python code that applies threshold tuning on model probabilities?

ML Python

import numpy as np
probs = np.array([0.2, 0.6, 0.8, 0.4, 0.9])
threshold = 0.7
preds = (probs >= threshold).astype(int)
print(preds.tolist())

A [0, 1, 0, 0, 1]

B [0, 0, 1, 0, 1]

C [0, 1, 1, 0, 1]

D [1, 1, 1, 0, 1]

💡 Hint

Check which probabilities are greater than or equal to 0.7.

Only 0.8 and 0.9 are >= 0.7, so their predictions are 1; others are 0.

❓ Model Choice

advanced

Choosing threshold for imbalanced data

You have a highly imbalanced dataset with very few positive cases. Which threshold tuning strategy is best to maximize recall while keeping false positives reasonable?

A Randomly select threshold to avoid bias

B Set a high threshold to reduce false positives, ignoring recall

C Use threshold 0.5 without tuning since it is default

D Set a low threshold to catch most positives, then use precision-recall curve to find balance

💡 Hint

Recall is about catching positives; think about how threshold affects it.

Lowering threshold increases recall but may increase false positives. Using precision-recall curve helps find a good trade-off.

❓ Metrics

advanced

Effect of threshold on F1 score

Given a model with these confusion matrix values at threshold 0.5: TP=40, FP=10, FN=20, TN=130. If threshold is increased to 0.7, TP=30, FP=5, FN=30, TN=135. What happens to the F1 score?

A F1 score decreases

B F1 score increases

C F1 score stays the same

D Cannot determine without precision and recall

💡 Hint

Calculate precision and recall for both thresholds, then compute F1.

At 0.5 threshold: precision=40/(40+10)=0.8, recall=40/(40+20)=0.67, F1≈0.73. At 0.7 threshold: precision=30/(30+5)=0.86, recall=30/(30+30)=0.5, F1≈0.63. F1 decreases.

🔧 Debug

expert

Debugging threshold tuning code with unexpected output

Consider this Python code snippet for threshold tuning. It produces unexpected predictions. What is the cause?

import numpy as np
probs = np.array([0.3, 0.7, 0.5, 0.9])
threshold = 0.5
preds = (probs > threshold).astype(int)
print(preds.tolist())

A Using '>' excludes probabilities equal to threshold, causing some positives to be missed

B The threshold variable is not used correctly; should be a list

C astype(int) converts floats incorrectly causing wrong predictions

D Numpy array probs is not sorted, causing wrong output

💡 Hint

Check how the comparison operator affects predictions equal to threshold.

The code uses '>' which excludes probabilities exactly equal to 0.5, so those are predicted negative. Using '>=' includes them.