Emerging Pattern Mining
Also known as: EP Mining, Contrast Pattern Mining, Differential Pattern Mining, Yükselen Örüntü Madenciliği
Emerging Pattern Mining (EPM) is a contrast-based data mining technique that identifies itemsets whose support increases significantly — or jumps from zero — when moving from one dataset (or class) to another. Introduced by Dong and Li in 1999, it is primarily used in classification, anomaly detection, and trend analysis tasks where discovering discriminative patterns between two populations or time periods is the central objective.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use Emerging Pattern Mining when you have two clearly separated datasets or classes and need interpretable, feature-level explanations for what distinguishes them — for example, comparing patient cohorts, network traffic snapshots, or customer segments across time. The method assumes discrete or discretized features and that a meaningful support threshold can be set. It is less suitable for high-dimensional continuous data without binning, very small datasets where support estimates are unreliable, or settings where black-box accuracy matters more than interpretability. Alternatives include decision trees for similar contrast tasks and association rule mining when no class label is available.
Strengths & limitations
- Produces highly interpretable, itemset-level rules that directly explain class differences
- Jumping Emerging Patterns provide unambiguous, class-exclusive discriminative features
- Border-based representation makes storage and enumeration tractable compared to naive search
- Naturally extends to multi-class and temporal trend discovery scenarios
- Requires discrete or pre-discretized features; continuous variables must be binned, introducing sensitivity to bin boundaries
- Worst-case search complexity is exponential in the number of items, limiting scalability to very wide datasets
- Performance depends heavily on the chosen growth-rate threshold rho and minimum support, which lack universal defaults
- Discovered patterns can be redundant; without post-pruning, the EP set may be very large and hard to inspect
Frequently asked
How does Emerging Pattern Mining differ from standard association rule mining?
Association rule mining finds frequently co-occurring itemsets in a single dataset without regard to class labels or contrast. Emerging Pattern Mining specifically measures how much an itemset's support grows between two datasets or classes, so it is inherently contrastive and supervised. This makes EPs directly useful for classification and anomaly detection, whereas association rules require additional filtering to extract class-discriminative knowledge.
What is a Jumping Emerging Pattern and why is it useful?
A Jumping Emerging Pattern is an itemset with zero support in one dataset and strictly positive support in another, giving it an infinite growth rate. Because it is entirely absent from the reference class, it carries unambiguous discriminative information. JEPs are especially valuable in classification because each one is a definitive marker of its associated class, making them easy to interpret and highly reliable when they match a test instance.
Can Emerging Pattern Mining handle more than two classes?
Yes. The binary two-dataset formulation extends naturally to multi-class settings by mining EPs for each class against the union of all other classes (one-vs-rest) or pairwise between every pair of classes. Each resulting EP set captures what distinguishes that class from the comparison group, and a scoring function aggregates votes across all EP sets when classifying new instances.
Sources
- Dong, G., & Li, J. (1999). Efficient mining of emerging patterns: Discovering trends and differences. ACM SIGKDD, 43–52. DOI: 10.1145/312129.312191 ↗
How to cite this page
ScholarGate. (2026, June 2). Emerging Pattern Mining. ScholarGate. https://scholargate.app/en/machine-learning/emerging-pattern-mining
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Association Rule MiningMachine learning↔ compare
- FP-GrowthMachine learning↔ compare
- Rule InductionMachine learning↔ compare