Classification models help solve problems like spam detection, disease diagnosis, and customer churn prediction. To judge how well your model works, you need evaluation metrics. These metrics highlight your model’s strengths and weaknesses, guide improvements, and ensure your results fit your specific needs.

Many users rely only on accuracy — the percentage of correct predictions. However, accuracy can be misleading, especially with imbalanced datasets. If just 1% of emails are spam, a model that always predicts "not spam" scores 99% accuracy but misses every real spam email. This shows why accuracy alone is not enough. You need a range of metrics to truly understand your model’s performance and the types of errors it makes.

Every binary classification prediction falls into one of four outcomes:

• True Positive (TP): Predicted positive and actually positive (e.g., flagged an email as spam, and it is spam);
• False Positive (FP): Predicted positive but actually negative (e.g., flagged an email as spam, but it is not spam);
• True Negative (TN): Predicted negative and actually negative (e.g., marked an email as not spam, and it is not spam);
• False Negative (FN): Predicted negative but actually positive (e.g., marked an email as not spam, but it is spam).

These outcomes are the building blocks for key metrics like precision, recall, and the confusion matrix. By examining TP, FP, TN, and FN, you gain a detailed view of your model’s prediction strengths and weaknesses, especially when different errors have different real-world impacts.