Market Basket Analysis

Affinity analysis in data mining is the search of stable groups of events that occur together in a certain subject area. It is based on the search for association rules which describe patterns of the relashionships between events.

In retail, this technique is used to perform market basket analysis, identifying stable sets of products acquired by supermarket customers in one purchase (for example, "potatoes, onions and salad", "pasta and ketchup", "beer and chips", "tea and baked goods", etc.). It allows optimizing product range and store layout to "encourage" buyers through cross-selling or up-selling.

The method is also successfully used in other areas, e.g., for studying web page visits, analysing and predicting telecommunication equipment failures, in medicine, etc.

This example demonstrates consumer basket analysis of a retail chain that sells household chemicals.

Launch demo

Download example

Algorithm Description

1. Data Import

a) Input data

The dataset contains information from 5,000 receipts. A receipt with a list of purchased items is considered a transaction, and each item in the receipt is an element of the transaction.

Name	Caption
id	ID
item	Item

2. Discovering association rules

To search for the association rules, we use the FP-growth algorithm.

The loaded transactions are fed into the Input Data Source port of the Association Rules node.

Interpretation of Results

To present the results, we use the Table visualizer which we set up for each port.

The Association rules table displays the sets of association rules and their indicators — support, confidence, and lift. This is the information that describes customer behavior. In the list obtained, we can see trivial patterns — for example, Fabric conditioner → Laundry detergent — as well as non-obvious ones (e.g., Paper towels → Air freshener).

The analysts should study each of the discovered rules and select the ones that are truly valuable.