Threshold tuning in ML Python - Model Metrics & Evaluation

Metrics & Evaluation - Threshold tuning

Which metric matters for Threshold tuning and WHY

Threshold tuning changes the cutoff point for deciding if a prediction is positive or negative. This affects Precision and Recall. We focus on these because changing the threshold shifts how many true positives, false positives, and false negatives we get. Accuracy alone can hide these changes. So, Precision and Recall help us understand the balance between catching positives and avoiding false alarms.

Confusion matrix example

      +-----------------------+
      |       Confusion Matrix |
      +-----------------------+
      |          | Predicted  |
      | Actual   | Pos | Neg |
      +----------+-----+-----+
      | Pos      | TP=80 | FN=20 |
      | Neg      | FP=10 | TN=90 |
      +-----------------------+

      Total samples = 80 + 20 + 10 + 90 = 200

      Precision = TP / (TP + FP) = 80 / (80 + 10) = 0.89
      Recall = TP / (TP + FN) = 80 / (80 + 20) = 0.80
    

Precision vs Recall tradeoff with examples

When you lower the threshold, the model predicts more positives:

• Recall increases: You catch more true positives (good for disease detection).
• Precision decreases: You get more false positives (more healthy people flagged).

When you raise the threshold, the model predicts fewer positives:

• Precision increases: Most predicted positives are correct (good for spam filters).
• Recall decreases: You miss some true positives.

Threshold tuning helps find the best balance for your problem.

Good vs Bad metric values for Threshold tuning

Good: Precision and Recall values close to each other and high (e.g., both above 0.8) show a balanced threshold.

Bad: Very high Precision but very low Recall means many positives are missed. Very high Recall but very low Precision means many false alarms.

Example: Precision=0.95 and Recall=0.30 is bad if missing positives is costly.

Common pitfalls in Threshold tuning metrics

• Ignoring class imbalance: Accuracy can be misleading if one class is much bigger.
• Overfitting threshold: Tuning threshold on test data leaks information and inflates performance.
• Using only one metric: Focusing only on Precision or Recall hides the full picture.
• Not validating threshold: Threshold should be chosen using validation data, not training or test data.

Self-check question

Your model has 98% accuracy but 12% recall on fraud detection. Is it good for production? Why or why not?

Answer: No, it is not good. The low recall (12%) means the model misses most fraud cases, which is dangerous. High accuracy is misleading because fraud is rare, so the model mostly predicts no fraud correctly but fails to catch fraud.

Key Result

Threshold tuning balances Precision and Recall by adjusting the cutoff, crucial for matching model behavior to real-world needs.

Practice

1. What is the main purpose of threshold tuning in machine learning classification?

easy

A. To find the best cutoff probability to decide between classes

B. To increase the size of the training dataset

C. To reduce the number of features used in the model

D. To speed up the training process

Solution

1. Step 1: Understand threshold tuning concept

   Threshold tuning is about choosing a cutoff value for predicted probabilities to decide class labels.

2. Step 2: Identify the main goal

   The goal is to find the cutoff that best separates positive and negative classes for better decisions.

3. Final Answer:

   To find the best cutoff probability to decide between classes -> Option A

4. Quick Check:

   Threshold tuning = best cutoff choice [OK]

Hint: Threshold tuning picks the cutoff to decide yes/no [OK]

Common Mistakes:

• Confusing threshold tuning with feature selection
• Thinking threshold tuning changes training data size
• Assuming threshold tuning speeds up training

2. Which of the following is the correct way to apply a threshold of 0.7 to predicted probabilities probs in Python to get binary predictions?

easy

A. preds = (probs > 0.7).astype(int)

B. preds = probs > 0.7

C. preds = int(probs > 0.7)

D. preds = probs >= 0.7

Solution

1. Step 1: Understand threshold application

   We compare each probability to 0.7 to get True/False, then convert to 0/1 integers.

2. Step 2: Check correct syntax

   Using (probs > 0.7).astype(int) converts boolean array to integer array correctly.

3. Final Answer:

   preds = (probs > 0.7).astype(int) -> Option A

4. Quick Check:

   Threshold applied with boolean then int cast [OK]

Hint: Use boolean comparison then convert to int for binary labels [OK]

Common Mistakes:

• Forgetting to convert boolean to int
• Using int() on entire array instead of element-wise
• Using >= instead of > changes threshold logic

3. Given the following code, what will be the printed F1 score after threshold tuning?

from sklearn.metrics import f1_score
probs = [0.2, 0.8, 0.6, 0.4]
true_labels = [0, 1, 1, 0]
threshold = 0.5
preds = [1 if p > threshold else 0 for p in probs]
f1 = f1_score(true_labels, preds)
print(round(f1, 2))

medium

A. 0.80

B. 0.67

C. 1.00

D. 0.50

Solution

1. Step 1: Calculate predictions with threshold 0.5

   probs > 0.5 gives preds = [0, 1, 1, 0]

2. Step 2: Compute F1 score for preds vs true_labels

   True positives = 2, false positives = 0, false negatives = 0, so F1 = 2*TP/(2*TP+FP+FN) = 2*2/(4+0+0) = 1.0, since preds and true_labels are identical.

3. Final Answer:

   1.00 -> Option C

4. Quick Check:

   Perfect match means F1 = 1.00 [OK]

Hint: Check predicted labels carefully before scoring [OK]

Common Mistakes:

• Miscomputing predictions from threshold
• Confusing precision and recall in F1 calculation
• Rounding errors in final score

4. The following code tries to tune threshold but gives an error. What is the error?

probs = [0.1, 0.4, 0.6, 0.9]
true_labels = [0, 0, 1, 1]
thresholds = [0.3, 0.5, 0.7]
best_f1 = 0
for t in thresholds:
    preds = (probs > t)
    f1 = f1_score(true_labels, preds)
    if f1 > best_f1:
        best_f1 = f1
print(best_f1)

medium

A. Thresholds list is empty

B. Missing import of f1_score

C. preds is boolean, should be integers

D. Loop variable t is not used

Solution

1. Step 1: Check code for missing imports

   The code uses f1_score but does not import it from sklearn.metrics.

2. Step 2: Identify error cause

   Without importing f1_score, Python will raise a NameError when calling f1_score.

3. Final Answer:

   Missing import of f1_score -> Option B

4. Quick Check:

   Always import functions before use [OK]

Hint: Check if all functions are imported before use [OK]

Common Mistakes:

• Assuming boolean preds cause error (they don't)
• Ignoring missing import errors
• Thinking loop variable is unused

5. You have a model predicting probabilities for a rare disease. You want to tune the threshold to catch as many sick patients as possible but avoid too many false alarms. Which approach best balances this trade-off?

hard

A. Choose threshold maximizing recall only

B. Choose threshold minimizing accuracy

C. Choose threshold maximizing precision only

D. Choose threshold maximizing F1 score

Solution

1. Step 1: Understand the trade-off

   High recall catches more sick patients but may increase false alarms; precision reduces false alarms but may miss sick patients.

2. Step 2: Identify best metric for balance

   F1 score balances precision and recall, making it best to tune threshold for this trade-off.

3. Final Answer:

   Choose threshold maximizing F1 score -> Option D

4. Quick Check:

   F1 balances recall and precision [OK]

Hint: Use F1 score to balance recall and precision [OK]

Common Mistakes:

• Maximizing recall ignores false alarms
• Maximizing precision ignores missed cases
• Minimizing accuracy is not meaningful