Now that we understand the concepts behind metrics like accuracy, precision, recall, F1-score, and the confusion matrix, let's see how to compute them efficiently using Scikit-learn's metrics module. This module provides optimized functions to evaluate your classification models based on the true labels and the predictions generated by your model.

Assuming you have already trained a classifier (like LogisticRegression, KNeighborsClassifier, or SVC as discussed earlier) and obtained predictions on your test set, you will typically have two arrays:

y_true: The ground truth labels for the test data.
y_pred: The labels predicted by your classifier for the test data.

Let's explore how to use these arrays to calculate the standard classification metrics.

Getting Started: Example Data

First, ensure you import the necessary functions from sklearn.metrics. For demonstration purposes, let's define some sample true labels and predicted labels. In a real scenario, these would come from your model's evaluation on a test set.

import numpy as np
from sklearn.metrics import (accuracy_score, confusion_matrix, precision_score,
                             recall_score, f1_score, classification_report)

# Example ground truth labels (binary classification)
y_true = np.array([1, 0, 1, 1, 0, 0, 1, 0, 0, 1])

# Example predicted labels from a hypothetical model
y_pred = np.array([1, 1, 1, 0, 0, 1, 1, 0, 0, 1])

# Example labels for multi-class classification
y_true_multi = np.array([0, 1, 2, 0, 1, 2, 0, 1, 2])
y_pred_multi = np.array([0, 2, 1, 0, 0, 2, 0, 1, 2])

Accuracy Score

Accuracy is the simplest metric, representing the proportion of correct predictions. It's calculated as:

\text{Accuracy} = \frac{\text{Number of Correct Predictions}}{\text{Total Number of Predictions}}

In Scikit-learn, you use the accuracy_score function:

# Calculate accuracy
acc = accuracy_score(y_true, y_pred)

print(f"Accuracy: {acc:.4f}")
# Expected Output: Accuracy: 0.7000

While easy to understand, remember accuracy can be misleading, especially for datasets with imbalanced classes.

Confusion Matrix

The confusion matrix provides a more detailed breakdown of prediction performance, showing counts of true negatives (TN), false positives (FP), false negatives (FN), and true positives (TP).

Use the confusion_matrix function:

# Calculate confusion matrix
cm = confusion_matrix(y_true, y_pred)

print("Confusion Matrix:")
print(cm)
# Expected Output:
# Confusion Matrix:
# [[3 2]
#  [1 4]]

The output is typically arranged as:

[[TN, FP],
 [FN, TP]]

So, in our example:

TN = 3 (Correctly predicted class 0)
FP = 2 (Incorrectly predicted class 1 when it was 0)
FN = 1 (Incorrectly predicted class 0 when it was 1)
TP = 4 (Correctly predicted class 1)

Visualizing the confusion matrix can often make it easier to interpret. Here's how you might generate a heatmap using Plotly:

Breakdown of correct and incorrect predictions for each class.

For multi-class problems, the matrix expands accordingly, showing counts for each actual vs. predicted class pair.

# Confusion matrix for multi-class example
cm_multi = confusion_matrix(y_true_multi, y_pred_multi)

print("\nMulti-class Confusion Matrix:")
print(cm_multi)
# Expected Output:
# Multi-class Confusion Matrix:
# [[3 0 0]
#  [1 1 1]
#  [0 1 2]]

Precision, Recall, and F1-Score

These metrics provide insights into specific aspects of performance, especially when class imbalance is a concern.

Precision: Measures the accuracy of positive predictions. $Precision = \frac{TP}{TP + FP}$ . Use precision_score.
Recall (Sensitivity): Measures how many actual positive cases were correctly identified. $Recall = \frac{TP}{TP + FN}$ . Use recall_score.
F1-Score: The harmonic mean of precision and recall, providing a single score that balances both. $F1 = 2 \times \frac{Precision \times Recall}{Precision + Recall}$ . Use f1_score.

# Calculate Precision, Recall, F1 for the positive class (label 1)
precision = precision_score(y_true, y_pred) # Default: pos_label=1
recall = recall_score(y_true, y_pred)     # Default: pos_label=1
f1 = f1_score(y_true, y_pred)             # Default: pos_label=1

print(f"\nBinary Classification Metrics (for class 1):")
print(f"Precision: {precision:.4f}") # TP / (TP + FP) = 4 / (4 + 2) = 0.6667
print(f"Recall:    {recall:.4f}")    # TP / (TP + FN) = 4 / (4 + 1) = 0.8000
print(f"F1-Score:  {f1:.4f}")       # 2 * (Prec * Rec) / (Prec + Rec) = 0.7273

# Expected Output:
# Binary Classification Metrics (for class 1):
# Precision: 0.6667
# Recall:    0.8000
# F1-Score:  0.7273

Handling Multi-class Metrics: For multi-class problems, you need to specify how to average these metrics across classes using the average parameter:

average='micro': Calculate metrics globally by counting total TP, FN, FP.
average='macro': Calculate metrics for each label, and find their unweighted mean. Does not take label imbalance into account.
average='weighted': Calculate metrics for each label, and find their average weighted by support (the number of true instances for each label). Accounts for label imbalance.
average=None: Returns the scores for each class individually.

# Calculate multi-class metrics with different averaging
precision_macro = precision_score(y_true_multi, y_pred_multi, average='macro')
recall_weighted = recall_score(y_true_multi, y_pred_multi, average='weighted')
f1_micro = f1_score(y_true_multi, y_pred_multi, average='micro')

print(f"\nMulti-class Metrics:")
print(f"Macro Precision:  {precision_macro:.4f}")
print(f"Weighted Recall:  {recall_weighted:.4f}")
print(f"Micro F1-Score:   {f1_micro:.4f}")

# Expected Output:
# Multi-class Metrics:
# Macro Precision:  0.6111
# Weighted Recall:  0.6667
# Micro F1-Score:   0.6667

Classification Report

Often, you'll want a summary of precision, recall, and F1-score for each class, along with the support (number of true instances per class). The classification_report function provides exactly this in a convenient text format.

# Generate the classification report for binary case
report_binary = classification_report(y_true, y_pred, target_names=['Class 0', 'Class 1'])
print("\nBinary Classification Report:")
print(report_binary)

# Expected Output:
# Binary Classification Report:
#               precision    recall  f1-score   support
#
#      Class 0       0.75      0.60      0.67         5
#      Class 1       0.67      0.80      0.73         5
#
#     accuracy                           0.70        10
#    macro avg       0.71      0.70      0.70        10
# weighted avg       0.71      0.70      0.70        10


# Generate the classification report for multi-class case
report_multi = classification_report(y_true_multi, y_pred_multi, target_names=['Class 0', 'Class 1', 'Class 2'])
print("\nMulti-class Classification Report:")
print(report_multi)

# Expected Output:
# Multi-class Classification Report:
#               precision    recall  f1-score   support
#
#      Class 0       0.75      1.00      0.86         3
#      Class 1       1.00      0.33      0.50         3
#      Class 2       0.67      0.67      0.67         3
#
#     accuracy                           0.67         9
#    macro avg       0.81      0.67      0.67         9
# weighted avg       0.81      0.67      0.67         9

The report includes:

Precision, recall, F1-score per class.
Support: The number of occurrences of each class in y_true.
Accuracy: Overall accuracy.
Macro Avg: Average of metrics per class (unweighted).
Weighted Avg: Average of metrics per class, weighted by support.

Scikit-learn's metrics module offers a straightforward way to quantify the performance of your classification models. Using functions like accuracy_score, confusion_matrix, precision_score, recall_score, f1_score, and the comprehensive classification_report allows you to move beyond simple accuracy and gain a deeper understanding of how your model behaves across different classes, which is essential for building effective classification systems.