Market Basket Analysis (Association Analysis)

Market basket analysis (also called association analysis) is a key method for uncovering relationships between items. It looks for combinations of items that frequently occur together in transactions. In other words, it helps retailers identify relationships between items that customers buy.

What does Association Analysis do?

Let's say you have set up your own online clothing store. Your goal is to achieve the highest possible revenue.

To achieve the highest possible sales, you want every customer to buy as much as possible. One way to encourage this is to suggest additional products. The key question is: Which product should you recommend? This is where market basket analysis or association analysis comes into play.

Market basket analysis answers the question: How likely is it that a customer buys product A if he or she already has product B in their basket?

Market basket analysis tells you which products are often bought together. So if a customer already has jeans and shoes in the shopping cart, how likely is it that they will also buy a shirt, socks, or a t-shirt?

Market Basket Analysis Example

To calculate a shopping cart analysis, you need a list of past purchases where you can see which products were bought together in one transaction.

You list the products and each row is a transaction. In this example, you have the products jeans, shirt, jacket, and shoes.

Each row is a transaction or purchase. "1" means bought, "0" means not bought. So the first person bought jeans, shirt, and shoes.

Now, to get results we can interpret, let's first calculate a market basket analysis using numiqo for this data. To do this, go to the market basket analysis calculator in numiqo and copy your data into the table.

Now we can specify a minimum support and a minimum confidence. For this data, numiqo returns these association rules:

The association rules are in the form: if the products under Lhs (left-hand side) are present in a transaction, then the products under Rhs (right-hand side) are also present with some probability.

Market Basket Analysis results interpretation

We look at the results of the basket analysis using the first set of association rules.

Frequency

The frequency in the results table tells us how often the products under Lhs and Rhs occur in a transaction. In our case, how often do shirt and shoes occur together?

So let's just count through how many transactions both occur in, which is 8 transactions.

Support

The support tells us what percentage of all transactions correspond to this one, or in other words, how likely it is that shirt and shoes will occur in a transaction. So we just divide the frequency by the number of all transactions.

There are 19 transactions in total, so we get 8/19, which is equal to 0.42. The probability of shirt and shoes occurring in a transaction is therefore 42 percent.

Confidence

Confidence tells us: if the products under Lhs are in an order, how likely is it that the products under Rhs are also in the shopping cart?

In our example this means: How likely is it that if shirt occurs in the cart, then shoes are also in the cart? We can calculate this by dividing the frequency of shirt and shoes by the frequency of shirt.

Lift

And finally, the lift. The lift indicates the factor by which the probability of buying\n the products under Rhs increases if the products under Lhs have already been bought. So, in our example, if the product shirt is in the shopping cart, it is 1.27 times more likely that shoes will be purchased than if shirt is not in the cart.

Market Basket Analysis and Data Mining

Shopping cart analysis is a method from the field of data mining. Depending on how much data is available, the analysis can be very computationally intensive.

However, with the Apriori algorithm, there are very effective methods to efficiently determine the association rules.

Critical Note on the Market Basket Analysis

Let's say your market basket analysis shows that if a person buys jeans and shoes, there is a high probability that they will also buy a shirt. Now you suggest a shirt to all customers who buy jeans and shoes. This increases the probability that a shirt will be bought under this condition, and a future market basket analysis will be biased.