How do you interpret and visualize the results of association rule mining in EDA?

1 Metrics for association rules

When conducting association rule mining, selecting the most relevant and interesting rules from a large set of possibilities is a challenge. To do this effectively, you should use metrics to measure the quality and significance of the rules. Common metrics include support, which shows the proportion of transactions containing both the antecedent and consequent; confidence, which indicates how reliable the rule is in predicting the consequent; lift, which shows how much the rule deviates from randomness; leverage, which measures how much the rule increases or decreases the occurrence of the consequent; and conviction, which indicates how much the rule depends on the consequent.

2 Filters for association rules

When applying association rule mining, it is necessary to reduce the number of rules to a manageable and meaningful subset. To do this, filters are often applied based on your goals and criteria. These filters include minimum support, confidence, and lift thresholds, as well as a maximum length limit. Additionally, domain knowledge can be used to filter out rules that are not meaningful or actionable for your specific context and objectives. By using these filters, you can eliminate rare, insignificant, weak, unreliable, and random chance rules.

3 Graphs for association rules

Visualizing and communicating the results of association rule mining is a third challenge. To do this, you can use graphs to display and compare the rules based on their metrics and filters. Scatter plots show the relationship between two metrics, such as support and confidence, for each rule. Matrix plots demonstrate the support and confidence of each rule as a color-coded cell in a matrix. Network graphs represent the rules as nodes and edges in a network, with nodes representing the items and edges representing the rules. Parallel coordinates plots illustrate the rules as lines in a parallel coordinates system, with each axis representing a metric, such as support, confidence, lift, or leverage. These graphs can help you identify patterns, outliers, trade-offs, connections, clusters, variations, and correlations among the items and rules.

4 Here’s what else to consider

This is a space to share examples, stories, or insights that don’t fit into any of the previous sections. What else would you like to add?