Model comparison strategies in ML Python - Model Metrics & Evaluation

Metrics & Evaluation - Model comparison strategies

Which metric matters for model comparison and WHY

When comparing models, the metric you choose depends on your goal. For example, if you want to know how often the model is correct overall, use accuracy. If you care about finding all positive cases, use recall. If you want to avoid false alarms, use precision. Sometimes, a balance like F1 score is best. For ranking or probability quality, use AUC. Choosing the right metric helps you pick the model that fits your real need.

Confusion matrix example for model comparison

      Actual \ Predicted | Positive | Negative
      -------------------|----------|---------
      Positive           |    TP=80 |   FN=20
      Negative           |    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
      Accuracy = (TP + TN) / Total = (80 + 90) / 200 = 0.85
      F1 Score = 2 * (Precision * Recall) / (Precision + Recall) ≈ 0.84
    

This matrix helps compare models by showing where they make mistakes.

Precision vs Recall tradeoff with examples

Imagine a spam email filter:

• High precision: Most emails marked as spam really are spam. Good to avoid losing important emails.
• High recall: Most spam emails are caught. Good to keep inbox clean.

Improving one often lowers the other. For example, marking fewer emails as spam raises precision but lowers recall.

Choosing which to prioritize depends on what matters more: missing spam or wrongly blocking good emails.

What good vs bad metric values look like

For a balanced classification task:

• Good model: Accuracy > 85%, Precision and Recall both > 80%, F1 score > 0.8
• Bad model: Accuracy near 50% (random guess), Precision or Recall very low (below 50%)

For imbalanced data, accuracy can be misleading. High recall with low precision or vice versa shows the model favors one side too much.

Common pitfalls in model comparison metrics

• Accuracy paradox: High accuracy but poor detection of minority class in imbalanced data.
• Data leakage: Using test data information during training inflates metrics falsely.
• Overfitting indicators: Very high training accuracy but low test accuracy means model memorizes data, not generalizes.
• Ignoring metric context: Using only one metric without considering the problem needs can mislead model choice.

Self-check question

Your model has 98% accuracy but only 12% recall on fraud cases. Is it good for production?

Answer: No. Even though accuracy is high, the model misses most fraud cases (low recall). For fraud detection, catching fraud (high recall) is critical. This model would let many frauds pass undetected.

Key Result

Choosing the right metric like precision, recall, or F1 score based on your goal is key to fair and useful model comparison.

Practice

1. What is the main purpose of model comparison in machine learning?

easy

A. To increase the size of the training data

B. To reduce the number of features in the dataset

C. To find the best model for the task using fair evaluation

D. To speed up the training process

Solution

1. Step 1: Understand model comparison goal

   Model comparison aims to select the best model by evaluating multiple models on the same task.

2. Step 2: Identify correct purpose

   Increasing data size, reducing features, or speeding training are not the main goals of model comparison.

3. Final Answer:

   To find the best model for the task using fair evaluation -> Option C

4. Quick Check:

   Model comparison = best model selection [OK]

Hint: Model comparison picks the best model fairly [OK]

Common Mistakes:

• Confusing model comparison with data preprocessing
• Thinking it speeds up training
• Mixing it with feature selection

2. Which of the following is the correct way to compare two models using accuracy in Python?

easy

A. print(accuracy_score(y_true, model1.predict(X)) > accuracy_score(y_true, model2.predict(X)))

B. print(model1.accuracy > model2.accuracy)

C. print(model1.predict(X) > model2.predict(X))

D. print(model1.score(X_train) == model2.score(X_test))

Solution

1. Step 1: Understand accuracy comparison

   Accuracy should be computed using true labels and predictions, not model attributes directly.

2. Step 2: Check each option

   print(accuracy_score(y_true, model1.predict(X)) > accuracy_score(y_true, model2.predict(X))) correctly uses accuracy_score with true labels and predictions for both models.

3. Final Answer:

   print(accuracy_score(y_true, model1.predict(X)) > accuracy_score(y_true, model2.predict(X))) -> Option A

4. Quick Check:

   Use accuracy_score with true labels and predictions [OK]

Hint: Compare metrics computed on true labels and predictions [OK]

Common Mistakes:

• Comparing model attributes directly
• Comparing predictions without labels
• Mixing train and test data in scoring

3. Given the code below, what will be printed?

from sklearn.model_selection import cross_val_score
from sklearn.linear_model import LogisticRegression

model1 = LogisticRegression()
model2 = LogisticRegression(C=0.1)

scores1 = cross_val_score(model1, X, y, cv=5)
scores2 = cross_val_score(model2, X, y, cv=5)

print(scores1.mean() > scores2.mean())

medium

A. False if model1 has higher average cross-validation score

B. True if model1 has higher average cross-validation score

C. Always True

D. SyntaxError due to missing imports

Solution

1. Step 1: Understand cross_val_score output

   cross_val_score returns an array of scores for each fold; mean() gives average performance.

2. Step 2: Compare mean scores

   The print statement compares which model has higher average score and prints True or False accordingly.

3. Final Answer:

   True if model1 has higher average cross-validation score -> Option B

4. Quick Check:

   Mean cross-val scores compared, prints True if model1 better [OK]

Hint: Cross-validation mean score comparison prints True if first is better [OK]

Common Mistakes:

• Assuming print always outputs True
• Confusing cross_val_score output with single score
• Missing that mean() is needed for average

4. You run this code to compare two models but get an error:

from sklearn.metrics import accuracy_score

acc1 = accuracy_score(y_true, model1.predict(X_test))
acc2 = accuracy_score(y_true, model2.predict(X_test))

if acc1 > acc2:
    print('Model 1 is better')
else
    print('Model 2 is better')

What is the error and how to fix it?

medium

A. IndentationError due to wrong indent; fix indentation

B. NameError because y_true is undefined; define y_true

C. TypeError because predict returns probabilities; use predict_classes

D. SyntaxError due to missing colon after else; add ':' after else

Solution

1. Step 1: Identify syntax error in if-else block

   The else statement is missing a colon at the end, causing a SyntaxError.

2. Step 2: Fix syntax by adding colon

   Add ':' after else to correct the syntax and run the code successfully.

3. Final Answer:

   SyntaxError due to missing colon after else; add ':' after else -> Option D

4. Quick Check:

   Missing colon after else causes SyntaxError [OK]

Hint: Check colons after if/else statements [OK]

Common Mistakes:

• Ignoring missing colon after else
• Assuming predict returns classes always
• Confusing indentation error with syntax error

5. You want to compare three models fairly on a small dataset. Which strategy is best to get reliable results?

hard

A. Use cross-validation with the same metric for all models

B. Train models on different datasets and compare their training loss

C. Use a single train/test split and compare accuracy

D. Compare models only by their training accuracy

Solution

1. Step 1: Understand fair comparison on small data

   Single splits can be unreliable; training accuracy is biased; different datasets cause unfairness.

2. Step 2: Choose cross-validation with same metric

   Cross-validation uses multiple splits to reduce randomness and comparing with the same metric ensures fairness.

3. Final Answer:

   Use cross-validation with the same metric for all models -> Option A

4. Quick Check:

   Cross-validation + same metric = reliable fair comparison [OK]

Hint: Cross-validation with same metric ensures fair comparison [OK]

Common Mistakes:

• Using single train/test split only
• Comparing training accuracy instead of test
• Using different datasets for each model