What Patterns Hide in Baskets
The viewer learns what association mining is and how transaction data can reveal items that tend to appear together.
Find the Hidden Patterns shows how association mining turns transaction data into clear links between items that often appear together. By the end, you'll know: what association mining means, how transaction data is read, and how item pairs reveal patterns. When people buy, click, or watch things, they leave behind transaction data. Association mining looks for items that show up together more often than you’d expect. That is how you find patterns in baskets, carts, and clickstreams. For example, if many shoppers who buy pasta also buy tomato sauce, that co-occurrence matters. You are not guessing from one case. You are scanning lots of records and asking, what tends to travel together again and again? Now let’s name the pieces. A transaction is one record, like one shopping trip or one session. An itemset is the group of items inside that record. If the same itemset keeps appearing across many records, it starts to matter. The data is usually categorical and binary for this task. Categorical means each item is either present or not present, not a number you average. Binary presence keeps the question simple: did this basket include bread, yes or no? Did this session include a search, yes or no? From there, you build a rule in if-then form. The left side is the antecedent, and the right side is the consequent. So you might read it as, if someone buys diapers, then they also buy wipes. Support, confidence, and lift tell you whether that rule is common, reliable, and genuinely interesting.
Turning Data Into Rules
The viewer learns how raw transactions become if-then rules and how support, confidence, and lift each help judge those rules.
So now that we know the parts, the next step is turning rows of transactions into rules you can actually read. Instead of staring at thousands of baskets, you want simple statements like, if X appears, Y often appears too. That shift matters because rules are easier to use than raw records. A store can place products together, a website can suggest the next click, and an analyst can explain the pattern without digging through every single transaction. Support is the first filter. It tells you how often an itemset appears across the full dataset. If a pattern shows up in only a tiny handful of records, it is usually too fragile to trust. Think of support as counting how many transactions contain the combination you care about. If 200 out of 1,000 baskets contain bread and butter together, that itemset has a support of 20 percent. That number tells you the pattern is not just a fluke. This is why support comes before deeper testing. It removes rare coincidences early, so you spend time on patterns that actually occur often enough to be useful. Confidence answers a different question. Once the left-hand side appears, how often does the right-hand side also appear? If you see diapers in a basket, confidence tells you how often wipes show up too. You calculate it by looking only at the transactions that contain the antecedent. If 80 out of 100 diaper baskets also include wipes, the confidence is 80 percent. That makes the rule feel dependable in the situations where it applies. But confidence has a limit. A rule can look strong just because the right-hand item is common everywhere. That is why confidence is useful, but not enough on its own. That is where lift comes in. Lift checks whether the rule is better than chance based on how common the right-hand item already is. It asks, are these two items really linked, or would the result happen a lot anyway? Say a product like bottled water is already in many baskets. If bread baskets often include bottled water, confidence may look high simply because bottled water is popular to begin with. Lift compares the rule against that baseline popularity, so you can see whether the pairing is actually special. A lift above 1 suggests the items occur together more than expected. Around 1 means the rule is mostly what you would predict from chance and commonness. Below 1 means the left side may even be associated with the absence of the right side. So lift helps you separate a real association from a busy-looking coincidence. It is the check that keeps you from promoting a rule just because one item shows up everywhere. When confidence says, ‘this often works,’ lift asks, ‘is it actually interesting?’
