Performance Metrics: Jaccard Index

What is the Jaccard Index?

The Jaccard Index (JI) is a performance metric that also has many other applications in a variety of fields and domains. That’s why it has a lot of names:

• Jaccard Coefficient
• Jaccard’s Coefficient of Community
• Jaccard Similarity Coefficient
• Jaccard–Tanimoto Coefficient
• Tanimoto Similarity Coefficient
• Short’s Measure
• Intersection over Union (see image)

The Jaccard Index can be calculated as follows:

In general, the JI is a proper tool for assessing the similarity and diversity of data sets. Within the context of evaluating a classifier, the JI can be interpreted as a measure of overlap between the ground truth and estimated classes, with a focus on true positives and ignoring true negatives.

The Jaccard Index is linearly related to the F-Measure:

Consequently, the JI suffers from the same criticism as the F-Measure: It exclusively focuses on the positive class and ignores the true negatives (TN). That’s why it is not the best performance metric in situations of class imbalance.

Further reading:

• The Jaccard Index in Text Mining
• The Jaccard Index in Genome Studies