Market Basket Analysis is the data mining method that allows the discovery of products purchased together at online and brick-mortar stores. However, when we have a long list of product combinations, an essential question is prioritizing them and learning which bundles are more critical. To solve this problem, we must ask: What is Lift in Market Basket Analysis?
Lift is the metric that helps us determine if combining a product with another improves the chances of making a sale. Furthermore, we can also discover if an association rule is not having any effect or, worse still, is detrimental.
In this post, we will explain what Lift in Market Basket Analysis is. Firstly, we will define Lift from a data mining perspective. Then, we will follow up with examples and how to use Lift to make decisions.
In this article:
Support Lift Confidence in Market Basket Analysis
Before explaining what Lift is, let’s look at two other indispensable metrics in market basket analysis: Support and confidence.
We examined these concepts in-depth in our previous blog post:
> What Is Market Basket Analysis And How It Can Increase Your Sales
Check it out before reading on.
Also, check out more of our content on business and technology:
- What Is Knowledge Management Strategy?
- Maximizing Workplace Safety Through Innovative Management Software Solutions
- The Evolution And Impact Of Human Capital Management
- 5 Ways To Avoid Phishing Attacks
- What Is AWS? Services, Costs And Benefits
Now, we discuss support, confidence and lift. Moving on!
Support
Support measures how frequently an association rule happens in a dataset.
Market Basket Analysis sets association rules around product bundles: For example, let’s say we are analyzing offering a sleeve with a laptop. We would define the association rule as follows:
IF [‘Laptop’] THEN [‘Laptop Sleeve’]
To measure Support, we must determine how many transactions contain the association rule and divide that by the total number of transactions. For instance, if we have 856 transactions in a given time and 34 of those include both a laptop and a laptop sleeve, then the Support for that association rule is 0,0397 (3,97%).
Measuring Support is essential to determine if an association rule significantly affects the overall number of purchases. Moreover, we can use Support to decide whether it is worth pursuing an association rule.
Confidence
Confidence measures how strong an association rule is. That is to say, in market basket analysis terms, how likely is a second product to be present in the basket if the first is. To calculate it, divide the number of purchases with both products (or group of products) over the number of transactions that include only the first product (or group), the antecedent part of the association rule.
Returning to our laptop sleeve example, we already know 34 baskets include laptops and sleeves. Now, let’s assume that 56 transactions of the 856 have a laptop, regardless of what other items are present. The rule’s confidence would be 34 divided by 56, which is 0.6071 (60.71%). As a result, if a transaction includes a laptop, it is 60.71% likely also to have a laptop sleeve.
Measuring confidence is critical to ascertain if a product combination is likely or unlikely to happen. Moreover, by comparing different product combinations’ confidences, you can decide which ones to choose to pursue marketing actions.
What Is Lift in Data Mining
In data mining and association rule learning, lift measures the performance of a targeting model (known as an association rule) at predicting a specific outcome compared with a random choice. Therefore, Lift is the ratio between target and average response. That is, Lift is a ratio between confidence and expected confidence.
While Market Basket Analysis is an example of association rule mining, Lift applies to data mining.
For example, assume that the probability of customers in a particular group canceling an online subscription is 20%. In comparison, the likelihood of any customer canceling is 5%, regardless of group. The lift ratio of the association rules defined by the customer as part of the group is 0.2 / 0.05, which equals 4.
As Lift is a ratio, it can have a value greater or below 1, depending on whether the model performs better than the average at predicting the outcome. As a result:
- If Lift is greater than 1, the target response is likelier than the average response. Therefore, the association rule improves the chances of the outcome.
- If Lift is below 1, the target response is less likely than the average response. As a result, the association rule lessens the chances of the desired outcome.
- A lift of 1 means that the model (or association rule) does not affect the outcome.
Lift in Market Basket Analysis
In Market Basket Analysis, expected confidence is the probability that the second product or group is in the basket regardless of any preconditions. That is to say, expected confidence is the number of purchases that include the second product divided by the total number of transactions.
Lift Data Mining Example
Let’s take a look at some examples:
Customer Cancelling Subscription
In the canceling subscription example, we established that the lift ratio equals 4. Since the Lift is greater than 1 in this case, the association rule (that the customer is part of a given group) increases the chances of canceling the subscription.
From a business perspective, we can take actions targeted at customers from that group to prevent it. For example, a live representative could call him to establish a close relationship, perform a survey or identify opportunities to serve them better.
Laptop and Laptop Sleeve
Returning to our Laptop and Sleeve example, we already established that the rule’s confidence is 60.71%. That is to say, 34 transactions include the laptop and the sleeve, and over 56 transactions have just the computer.
Let’s assume that 88 purchases include the Laptop Sleeve. As a result, the expected confidence would be 88 over 856, which results in 0.1028 (10.28%).
Do you believe that placing the laptop and the sleeve together improves the customer’s chance of buying both items?
Lift, which is confidence divided by expected confidence, results in 60.71% divided by 10.28%, which results in a lift ratio of 5.91. Thus, buying a laptop improves the chance of buying a laptop sleeve.
Customer Buying a Smartphone and Its Case
Now let’s consider another case, this one from Market Basket Analysis:
Suppose the probability of buying an Android Smartphone and a Smartphone case in the same basket is 9%. In contrast, the likelihood of purchasing the Smartphone case is 4% (regardless of any other items in the basket). The lift ratio would be 2.25. Therefore, buying a Smartphone improves the chances of also buying a case.
With this knowledge, a business could offer coupons or discounts targeted at customers buying a smartphone. Also, if we have a brick-mortar store, we can offer both items in bundles or place them close to one another. In online stores, we could include a section for “customers who bought an item also bought…” to facilitate purchasing.
Market Basket Analysis in Tableau
We wrote a post about how to use Tableau to perform a Market Basket Analysis. Follow the link:
> How to Conduct a Market Basket Analysis in Tableau
Lift Formula Data Mining
To sum it all up, the lift formula is as follows: Given an event A and an event B, the Lift of both events is:
Lift A and B = Confidence A and B / Expected Confidence (Confidence of B)
That is to say:
Lift A and B = (Number of Occurrences of A and B / Total Number of Events)
Divided by
number of occurrences of B (Regardless of other factors) / Total Number of Events
This Lift formula between two products or events is also known as Two Way Lift.
Using Market Basket Analysis Lift To Make Decisions
We can use Lift, Support and Confidence measures to determine the strongest association rules between products in a market basket. Therefore, we can rank them and prioritize marketing actions around them.
Let’s take a look at another example: Assume that after performing a Market Basket Analysis, we came up with the following results table:
What marketing decisions would you make based on these outcomes?
Let’s see:
- The Laptop / Laptop sleeve and Headphones / Bluetooth adapter products combinations have good values. Lift is high in both instances, and Support for both rules is significant.
- The Smartphone / Smartphone Case bundle has high confidence and lift. However, Support is somewhat low. As a result, it might not be worth pursuing marketing actions around this rule. That is to say, there is a high chance of these products selling together. However, you don’t sell a lot of these products, regardless.
- The Home bedding set and pillow combination have good Support and confidence. However, Lift is just below 1. You are selling a good number of home bedding sets and pillows. Nevertheless, bundling them together doesn’t have any effect on how many you would sell.
Other Data Mining Examples
Beyond lift and market basket analysis, there are other data mining examples. Check out this article: 15 Examples Of Data Mining In Real Life: Uncovering Hidden Insights.
Begin Prioritizing Association Rules Using Lift in Market Basket Analysis
This post explained Lift in market basket analysis and how to calculate it. Moreover, we presented practical examples to set you up and apply these practices yourself.
We invite you to take action now. Market basket analysis allows you to single out products purchased in bundles and define activities around them. As a result, you have the potential to increase your sales and improve your customer’s experience at the same time.
What are your thoughts on what is lift in market basket analysis? Have you measured Lift? Please share your experience in the comments section.
Check Out More of Our Content
While you are here, check out our latest articles about cloud computing, cybersecurity, cloud security and more:
