Market Basket Analysis (MBA) is primarily concerned with establishing which products tend to be purchased together (or not purchased together) so that cross-selling activities, store layout, marketing activities and so on can be optimized. It is usually a specialized form of association rule mining (ARM) which seeks to find rules that express how customers behave. The general format of a rule is:
IF (these products) THEN (those products)
Mining large transaction databases tends to produce an excess of rules – most of them trivial. To sort the wheat from the chaff a number of measures of interestingness are used, the most popular being support (what fraction of the database contains the products in the first part of the rule) and confidence (what percentage of occurrences of ‘these products’ is associated with ‘those products’). These two measures make sure that a rule is derived from a large number of transactions, and that it has a high level of accuracy. Placing thresholds on these two measures (80% confidence say and 1% support) reduces the number of rules that need to be considered considerably.
Predictive MBA is concerned with transactions that are sequenced through time. It might be that customers tend to buy one product and then purchase other products soon after. The applications are obvious and with the rise of predictive analytics this is one of the more recent MBA innovations.
A large number of suppliers provide MBA software, some of it expensive and complex and some newer entrants with low cost, on-demand solutions. 11Ants Insight Portal – low cost on-line service, with surprisingly good analytics. IBM SPSS suite includes basket analysis as do the suites from Angoss and SAS. Many other options are available, and any data mining tool-set with ARM will support MBA activity.