Machine learning models are only as good as the metrics used to evaluate them. When building a classification model, three evaluation metrics often cause confusion: accuracy, precision, and recall. Although they seem similar, they measure very different aspects of model performance — and choosing the wrong one can lead to dangerously misleading conclusions.

What Are Accuracy, Precision, and Recall?

These are classification evaluation metrics used to measure how well a model performs. Each answers a different question about your model's behaviour.

All three are calculated from a confusion matrix — the foundation of classification evaluation.

Understanding the Confusion Matrix

A confusion matrix is a 2×2 table that breaks down model predictions into four categories. Every classification metric derives from these four values.

All Formulas at a Glance

Accuracy Explained

Accuracy measures the proportion of all correct predictions out of total predictions — both positive and negative.

When Accuracy Works Well

• Classes are roughly balanced (e.g. 50/50 split)
• Cost of false positives and false negatives is similar
• You need a quick, broad overview of model performance

Precision Explained

Precision answers: "Of all the cases my model flagged as positive, how many actually were?" It penalises false alarms.

When to Use Precision

Use precision when false positives are costly. In fraud detection, wrongly flagging a legitimate transaction creates customer frustration and support overhead. High precision ensures flagged transactions are genuinely suspicious.

Recall Explained

Recall answers: "Of all the actual positives that existed, how many did my model catch?" It penalises missed cases.

When to Use Recall

Use recall when false negatives are dangerous. In cancer screening, missing a true positive case is far worse than an unnecessary follow-up test. Maximising recall ensures the model catches as many real cases as possible.

Worked Example — Spam Detection

Suppose a classifier evaluates 200 emails and produces the following confusion matrix values. Let's calculate all four metrics:

// in this balanced example all three metrics agree. The dangerous cases are imbalanced datasets — covered next.

The Accuracy Paradox

In imbalanced datasets, accuracy becomes dangerously misleading. A model can achieve very high accuracy by simply predicting the majority class every time — while completely failing at its actual job.

The Precision–Recall Trade-Off

There is an inherent tension between precision and recall. Lowering the classification threshold catches more positives (higher recall) but also more false alarms (lower precision). Raising the threshold reduces false alarms (higher precision) but misses more real positives (lower recall).

// as the threshold rises, precision climbs and recall falls. F1 score peaks where the two balance — the optimal zone.

What Is the F1 Score and Why It Matters

The F1 score is the harmonic mean of precision and recall. It gives a single number that balances both, making it the go-to metric for imbalanced datasets where you care about both false positives and false negatives.

• F1 = 1.0 — perfect precision and recall
• F1 = 0.0 — either precision or recall is zero
• F1 > 0.7 — generally considered good on imbalanced data

Metrics for Imbalanced Datasets

In fraud detection where only 1% of transactions are fraudulent, a model that predicts "not fraud" for everything achieves 99% accuracy but recall of 0%. The imbalance exposes accuracy as useless.

Which Metric Should You Use?

Always choose your evaluation metric based on the business consequences of each type of error in your specific problem.

Python Implementation

Scikit-learn makes computing all four metrics trivial. Here is a complete example on a binary classification problem:

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

# ── Example predictions ──────────────────────────────────
y_true = [1, 0, 1, 1, 0, 1, 0, 0, 1, 1]   # actual labels
y_pred = [1, 0, 1, 0, 0, 1, 1, 0, 1, 0]   # model predictions

# ── Individual metrics ───────────────────────────────────
print(f"Accuracy : {accuracy_score(y_true, y_pred):.4f}")
print(f"Precision: {precision_score(y_true, y_pred):.4f}")
print(f"Recall   : {recall_score(y_true, y_pred):.4f}")
print(f"F1 Score : {f1_score(y_true, y_pred):.4f}")

# ── Confusion matrix ─────────────────────────────────────
cm = confusion_matrix(y_true, y_pred)
print("\nConfusion Matrix:")
print(cm)
# [[TN, FP],
#  [FN, TP]]

# ── Full report (easiest option) ─────────────────────────
print("\nClassification Report:")
print(classification_report(y_true, y_pred))

# ── For imbalanced datasets — adjust averaging ───────────
# zero_division=0 avoids warnings when a class has no predictions
f1_macro  = f1_score(y_true, y_pred, average='macro')
f1_weighted = f1_score(y_true, y_pred, average='weighted')

// use classification_report() as your default — it prints precision, recall, F1, and support for every class in one call.

FAQs

Final Thoughts

Understanding accuracy vs precision vs recall is essential for building reliable machine learning models. Accuracy gives a broad overview, but precision and recall provide deeper insight into where your model fails and why.

• On balanced datasets with equal error costs — report accuracy
• When false positives are costly — optimise precision
• When false negatives are dangerous — optimise recall
• On imbalanced datasets — always use F1 score (and consider PR-AUC)

Always align your evaluation metric with the real-world consequences of each error type. The best model is not the one with the highest accuracy — it is the one that minimises the errors that matter most to your problem.