In words
What it is, why it matters, and what it is like.
Why am I learning this?
This technique allows you to discover hidden connections in lists of items grouped together, such as which products are frequently bought in the same shopping cart, or which symptoms often appear in the same patient record. Understanding these co-occurrences is essential for building recommendation engines that suggest related items, or for identifying significant risk factors in medical data where specific conditions cluster together. It provides the foundational logic for systems that predict what you might need next based on what you have already selected.
The idea, in plain terms
Imagine you manage a small grocery store and want to know which items are commonly purchased together. You have a ledger of every customer’s shopping basket. To find these patterns, you first look at individual items. If you set a rule that an item must appear in at least 10% of all baskets to be considered common, you can quickly identify the 'frequent items'. For example, if bread appears in 20% of baskets and milk appears in 15%, both are frequent. However, if sourdough flour appears in only 2% of baskets, it is rare. Any group containing sourdough flour (like a basket with sourdough flour and butter) cannot be more common than the sourdough flour itself, so you can ignore those groups entirely. This is the key efficiency: if a single ingredient is too rare, any larger shopping cart containing it must also be too rare. You only then check combinations of the frequent items, such as pairs like 'bread and milk'. If you find that 'bread and milk' appears together in 8% of baskets, it meets your threshold. This process scales because by eliminating rare ingredients early, you avoid checking millions of impossible combinations.
An analogy
Think of planning a dinner party where you want to know which dishes are typically served together. You look at a history of past meals and set a rule: a dish is 'essential' if it appears in at least 20% of the meals. If 'salad' never appears, no meal containing salad can be considered a common pattern, so you stop considering any combination with salad. You then focus only on the essential dishes, like 'pasta' and 'wine', checking how often they appear together. This works because the presence of the rare dish is the bottleneck; if it’s missing, the whole combination is rare. The analogy breaks down slightly because in a real database, items are counted per transaction, whereas in your head, you might imagine ingredients mixing into one large meal rather than separate events.
Definition
Apriori is a method for finding groups of items that appear together frequently in a dataset by first identifying individual common items and then systematically combining them, while ignoring any group containing an item that did not meet the minimum frequency requirement.
Where this sits
This technique is part of discovering structure without pre-labeled outcomes, a core goal of Unsupervised Learning. It relies on the same basic counting logic used in simple statistics but applies it to groups. It connects directly to Association Rule Mining, which uses these frequent groups to generate rules like 'if X is bought, Y is likely bought', by measuring how often the second item appears when the first is present (confidence) and whether this relationship is stronger than chance (lift).
