Online Association Rules
Online (Incremental) Association Rule Mining · Also known as: Incremental association rule mining, Streaming association rules, Online ARM, Incremental ARM
Online association rule mining discovers if-then patterns (e.g., buying bread implies buying butter) from transactional data that arrives incrementally or as a stream, updating existing rules and item counts without re-scanning the entire historical database each time new records arrive.
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 online association rule mining when transactional or event data arrives continuously or in frequent batches and rules must stay current without costly full re-scans — for example, real-time retail basket analysis, clickstream pattern tracking, or network intrusion detection on event logs. It is also appropriate when storage constraints prevent retaining the full history needed for a batch Apriori or FP-growth pass. Do not use it when the dataset is static and fits comfortably in memory — standard batch Apriori or FP-growth will produce equivalent results faster and with simpler code. Avoid it when the concept distribution shifts dramatically over time without a forgetting mechanism, as stale counts can mislead rule quality estimates.
Strengths & limitations
- Avoids full re-scans of historical data when new transactions arrive, dramatically reducing compute time.
- Produces continuously up-to-date rules suitable for real-time recommendation or alerting.
- Handles database insertions and, in some variants, deletions without restarting the mining process.
- Memory-efficient relative to buffering an ever-growing dataset for repeated batch runs.
- Composable with sliding-window or time-decay mechanisms to prioritise recent patterns.
- Maintaining accurate support counts across a large, evolving candidate set can still be memory-intensive.
- Without a forgetting mechanism, counts accumulate indefinitely, letting old patterns crowd out emerging ones.
- Implementation complexity is substantially higher than batch Apriori or FP-growth.
- Choosing minimum support and confidence thresholds remains manual and dataset-specific.
- Candidate itemset explosion is still possible if the transaction space is very wide.
Frequently asked
How is online association rule mining different from simply re-running Apriori on each new batch?
Re-running Apriori from scratch reads the entire historical database every time, which is O(n) in history size. Online ARM maintains running counts and updates only the affected entries, making each update proportional to the size of the new batch rather than total history.
What happens to rules when data is deleted from the database?
Algorithms that support decremental updates (such as FUP extensions) decrement counts for affected itemsets. Rules whose support falls below the minimum threshold after deletion are invalidated and removed.
How do I prevent the rule set from being dominated by old, irrelevant patterns?
Apply a sliding window (count only the last N transactions) or an exponential time-decay to counts so that recent transactions have higher weight. This forces old patterns to fade naturally as new data arrives.
Can online association rule mining handle concept drift?
Basic online ARM does not adapt to concept drift automatically. Adding a time-decay weight or a fixed sliding window introduces a form of forgetting that lets the rule set track shifting patterns, but explicit drift-detection hooks are needed for rapid, abrupt shifts.
What are reasonable starting thresholds for minimum support and confidence?
There is no universal answer; thresholds depend heavily on domain and transaction density. A common practice is to start with minimum support of 1–5% and minimum confidence of 50–70%, then adjust based on the volume and utility of discovered rules.
Sources
- Cheung, D. W., Han, J., Ng, V. T., & Wong, C. Y. (1996). Maintenance of discovered association rules in large databases: an incremental updating technique. In Proceedings of the 12th International Conference on Data Engineering (ICDE 1996), pp. 106–114. IEEE. link ↗
- Association rule learning. Wikipedia. link ↗
How to cite this page
ScholarGate. (2026, June 3). Online (Incremental) Association Rule Mining. ScholarGate. https://scholargate.app/en/machine-learning/online-association-rules
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.
- Apriori AlgorithmMachine learning↔ compare
- Association RulesMachine learning↔ compare
- FP-GrowthMachine learning↔ compare
- Online LearningMachine learning↔ compare
- Semi-supervised Association RulesMachine learning↔ compare