Let's say you own a grocery store Al's Local Mart (ALM). You want to increase sales. You read that big data can bring big sales. You are willing to re-shuffle product placement in your grocery store to do that. But how should products be re-shuffled to increase sales.
One way to re-shuffle product is put items close to each other - items that are often purchased together. A common example is cereal and milk. Now you want to implement this for other products.
You start to collect data on grocery sales in your mart. You write down on a piece of paper what was bought : (M)ilk, (O)ranges, (N)ectarines, (K)ola, (E)thiopian Coffee, (Y)ellow Banana.
 | ||
| Hand implementation of the Apriori Algorithm to find association of items in Market Basket Analysis. |
In addition to the raw data of what was bought together, you will also need to define what Minimum Support Threshold to use to create the recommendation system. The higher the number, the more obvious the association has to be for it to be recognized.
It the first table on the right, the number of times an item is purchased is listed. Those that don't occur often (in accordance with the threshold) are eliminated. I made this number high so that I can reduce the number of hand iterations. In the 2nd table on the right, combinations of items are tallied, and those below the threshold are eliminated. The last table factorizes all of the remaining associations. The most common association is : (O)ranges, (K)ola, and (E)thiopian Coffee are usually bought together.
As a grocer, you can now place (O)range, (K)ola, (E)thiopian Coffee close together. Or you can make a recommendation at the check out counter. Either way, your chance of increasing sales is good.
