Negative patterns are patterns whose components exhibit negative correlated behavior, while rare patterns occur infrequently but are of particular interest. An example includes:
In a jewelry data sale, sales of gold watches are infrequent though the patterns concerning the sale of gold watches could be interesting. In open market data, when we find that customers mostly buy Diet Coke or Coca-Cola Classic but not both, then buying Diet Coke and Coca-Cola Classic together can be termed a negative pattern.
A rare pattern has frequency support that is below the user-specified minimum support threshold. Since the occurrence frequencies of most of the itemsets are frequently below the minimum support threshold, it is significant in practice for the users to postulate for the irregular pattern. For example, if we need to find ways with at least one item costing over $700, we should specify such a restraint.
A negative pattern can be well understood using a null-transaction problem. This is whereby when there are high null transactions in a set of data. Example: Suppose a sewing store sells needle packages X and Y. It sells 200 packages each of X and Y. Still, only one transaction has both X and Y. Intuitively, X will be negatively correlated with Y because the purchase of one does not seem to inspire the buying of the other.
Therefore, the number of null transactions apart from the observed patterns might powerfully affect a measure’s assessment of whether a pattern is negatively correlated.