Support
When people look for patterns in shopping baskets, website clicks, or sets of symptoms, one of the first questions is simple: how frequently does this combination actually appear? In association rule learning, that frequency is called support, and it acts like a reality check for whether a pattern is common enough to matter.
What support means
Support is the proportion of transactions in a dataset that contain a given itemset. If 200 out of 1,000 receipts include both bread and butter, the support of {bread, butter} is 0.20, or 20%. For a rule such as {bread} → {butter}, support refers to how many transactions contain both the left-hand side and the right-hand side together. This is different from confidence, which asks: among bread purchases, how many also include butter? Support answers a more basic question: how common is this pattern in the full dataset?
Why it matters
Support is crucial because association rule mining can generate huge numbers of possible item combinations. Most are too rare to be useful. A minimum support threshold filters out infrequent patterns early, making algorithms like Apriori and FP-Growth practical.
- In retail, it prevents attention from going to item pairs that happened only a handful of times.
- In fraud analysis, low support can signal unusual behavior, but for rule discovery it usually means the pattern is too weak to generalize.
- In libraries such as mlxtend.frequent_patterns, support is a core output and filtering criterion.
Practical intuition
A rule with high confidence but very low support can be misleading: it sounds strong, but it barely occurs. Support keeps discovered rules grounded in enough data to be trustworthy and operationally useful. It is one of the main measures that decides which patterns survive the mining process and which are dismissed as statistical noise.
Support is the proportion or count of transactions in a dataset that contain a given itemset, or both the antecedent and consequent of an association rule. It measures how frequently a pattern occurs in the data. Support matters because it filters out rare, unreliable rules and defines which itemsets are frequent enough to mine efficiently, making association rule discovery computationally feasible and practically useful.
Imagine a supermarket looking through thousands of receipts to see which item combinations show up again and again. Support is simply the share of transactions that contain a particular item or group of items. If bread and milk appear together in 20 out of 100 baskets, their support is 20%.
In AI and machine learning, especially when finding shopping-style patterns in data, support answers a basic question: “How common is this pattern?” That matters because a pattern that appears only once or twice may be a coincidence, while one with high support is more likely to be useful and trustworthy. It helps systems focus on patterns that are actually common in the real data.