Note

If you encounter an error with the Jupyter Notebook on GitHub, please use this link to open it nbviewer

Basket Analysis

Introduction

Market Basket Analysis is a data mining technique employed to discover relationships and patterns within large datasets, particularly in the context of market analysis. By identifying frequently co-occurring items in transactions, businesses can gain valuable insights into customer behavior, optimize product placement, and enhance overall marketing strategies.

Data Understanding

This Online Retail II data set contains all the transactions occurring for a UK-based and registered, non-store online retail between 01/12/2009 and 09/12/2011. The company mainly sells unique all-occasion giftware. Many customers of the company are wholesalers.

The dataset can be accessed at the following link: https://www.kaggle.com/datasets/mashlyn/online-retail-ii-uci/

Business Goals

Market business analysis aims to provide businesses with a comprehensive understanding of their market environment, competition, opportunities, threats, and customer behavior to inform strategic decision-making.

Objectives

1. Association Rule Discovery: Identify associations and correlations among products or items in a dataset. Discover rules that indicate the likelihood of certain items being bought together.
2. Cross-Selling Opportunities: Uncover opportunities for cross-selling by understanding which products are frequently purchased together.
3. Optimizing Product Layout: Arrange products in-store or online in a way that encourages the purchase of related items, creating a more convenient and satisfying shopping experience.

Method

Apriori algorithm

• We utilize the Apriori algorithm for conducting market basket analysis.
• This algorithm proves highly effective in discerning frequent itemsets and deriving association rules, relying on predefined metrics like support and confidence.
• To execute the Apriori algorithm, we utilize the mlxtend library, a reliable Python library for machine learning extensions.

The following parameters are configured for the algorithm:

• Maximum Combination Length
  • We set the maximum combination length to 2 items.
  • This choice is made to focus on pairs of items, allowing for a more targeted analysis of co-occurrences.
• Minimum Occurrence Product
  • We set the minimum support threshold at 1% to focus on the most commonly purchased products.
• Minimum Co-Occurrence Support Threshold
  • A minimum co-occurrence support threshold of 1% is established to filter out infrequent itemsets.
  • This ensures that only associations with a significant presence in the dataset are considered.

Results and Analysis

Note

For a comprehensive and interactive view please visit the following link https://basket-analysis.streamlit.app/

Top 5 Products based on support

Network Profile

Using the Echarts network, we perform profiling to identify which groups of products are frequently purchased together. For the grouping method, we use nx.community.greedy_modularity_communities.

Bundling Products with a Minimum Mean Confidence of 50%

In this setting, with the mean confidence threshold at 50%, products like 'WHITE HANGING HEART T-LIGHT HOLDER' and 'RED HANGING HEART T-LIGHT HOLDER', despite having high support, did not appear as a result of the cross-selling products algorithm. This is because we aim to identify pairs or groups of products that are commonly purchased together, rather than individual popular items or items that are generally popular.

Product Heatmap Layout

Selected category: "RED STRIPE CERAMIC DRAWER KNOB"
The heatmap clearly indicates that "RED STRIPE" is the most interconnected product, showing strong support values with several other products. The implications are we need to put a RED STRIPE on the upper shelf and bundle it with other products at a special price.