View source for Complementary metrics

= Complementary Metrics in Machine Learning =

'''Complementary Metrics''' refer to pairs or groups of evaluation metrics that together provide a more complete and balanced understanding of a classification model’s performance. Because no single metric is perfect, especially in real-world and imbalanced datasets, these metrics are used together to highlight different strengths and weaknesses of a model.

== Why Use Complementary Metrics? ==

Using only one metric like [[Accuracy]] can be misleading — especially when dealing with imbalanced classes. Complementary metrics help you:

* Understand different types of errors (false positives vs false negatives)
* Choose a model that fits your specific use case
* Balance trade-offs (e.g., sensitivity vs specificity)

== Common Complementary Pairs ==

=== 1. Precision and Recall ===

* '''Precision''' focuses on how many predicted positives are correct.
* '''Recall''' (or [[Sensitivity]]) focuses on how many actual positives were caught.
* Complement each other: high precision may come with low recall, and vice versa.

→ Combined using the [[F1 Score]]

=== 2. Sensitivity and Specificity ===

* '''Sensitivity''' (Recall) = True Positive Rate
* '''Specificity''' = True Negative Rate
* Complement each other in binary classification tasks.

Example: In medical diagnosis,
* High Sensitivity ensures sick patients are detected.
* High Specificity ensures healthy people aren't misdiagnosed.

=== 3. Accuracy and F1 Score ===

* '''Accuracy''' is good for balanced datasets.
* '''F1 Score''' is better for imbalanced data where false negatives or positives matter more.

Together, they offer a more well-rounded picture.

=== 4. ROC and AUC ===

* The [[ROC Curve]] plots Sensitivity vs. 1 − Specificity.
* The AUC (Area Under the Curve) summarizes the ROC into a single score between 0 and 1.

→ These complement threshold-based metrics by offering a threshold-independent evaluation.

== Real-World Example ==

In a **spam detection system**:

* [[Precision]] tells you how many flagged emails are actually spam (important for avoiding loss of important emails).
* [[Recall]] tells you how many spam emails the system successfully detected.
* [[F1 Score]] balances the two.

== When to Use Complementary Metrics ==

* Your dataset is imbalanced
* You're working in high-risk domains (medicine, finance, law)
* You want a holistic view of model performance
* Model decisions have real-world consequences

== Visual Tools ==

* [[Confusion Matrix]]: Base for calculating most metrics
* [[ROC Curve]]: Visualizes trade-offs between Sensitivity and Specificity

== Related Pages ==

* [[Precision]]
* [[Recall]]
* [[Specificity]]
* [[F1 Score]]
* [[Accuracy]]
* [[ROC Curve]]
* [[Confusion Matrix]]

== SEO Keywords ==

complementary metrics in machine learning, precision vs recall, sensitivity vs specificity, model evaluation strategies, performance metrics comparison, balanced evaluation, f1 vs accuracy, ROC AUC evaluation
[[Category:Artificial Intelligence]]