CUT MIND - Reglas de Asociación

Introduction to Association Rules in Data Mining

Overview of Association Rules

• The video introduces association rules, a key application in data mining, particularly for analyzing product purchases in stores.

• It highlights the potential to discover patterns that indicate which products are frequently bought together or influenced by other products.

Applications of Association Analysis

• Various applications include product arrangement, defining navigation patterns within stores, suggesting effective promotions for product pairs, and generating specific discounts for customers.

• The analysis can also be applied to survey data, treating responses as purchases to uncover hidden patterns and relationships among variables.

Understanding Key Concepts

Definitions Related to Association Rules

• The most common algorithm used is the Apriori algorithm, which automatically finds association rules from data.

• An itemset is defined as a collection of one or more items; for example, "milk," "diapers," and "beer" form an itemset.

Support and Frequent Itemsets

• Support refers to how often an itemset appears in the database. A relative support is calculated by dividing the number of occurrences by total transactions.

• An itemset is considered frequent if its support exceeds a predefined threshold.

Interpreting Association Rules

Structure of Association Rules

• An association rule takes the form X → Y (e.g., milk and diapers imply beer), indicating that purchasing X suggests purchasing Y.

• It's important to note that this implication relies on empirical confidence and support values derived from data.

Example with Transaction Data

• A table illustrates transaction data where each row represents a purchase with listed products.

• Different combinations yield distinct rules despite involving the same items (e.g., diaper purchase implying milk vs. beer).

Summary of Key Learnings

Recap on Association Rules

• The video summarizes that association rules help analyze purchase data from various contexts including surveys.

• Important definitions covered include itemsets, support, and association rules themselves.

Performance Indicators Behind Association Rules

Evaluating Credibility of Decisions

• The next segment focuses on understanding performance indicators related to association rules for assessing decision credibility based on found rules.

Calculating Support and Confidence

• Support measures frequency relative to transactions; e.g., calculating support for "milk, diapers → beer" involves counting relevant transactions against total ones.

Understanding Confidence and Lift in Association Rules

Calculating Occurrences and Probabilities

• The calculation of occurrences involves identifying the number of rows containing specific items, such as milk and diapers, which are subsets of a larger set that includes beer. This is crucial for determining probabilities.

• The relationship between calculating confidence for an association rule (e.g., milk → beer) and conditional probability is highlighted. Conditional probability can be derived from joint probabilities divided by the probability of the antecedent item.

Interpreting Confidence Values

• A confidence value of 0.67 indicates that 67% of consumers who bought milk and diapers also purchased beer. High confidence values may not always provide useful information if the consequent item's support is already high independently.

• For instance, if the prior probability of buying beer is also 70%, knowing that a customer bought milk does not significantly change the likelihood of them purchasing beer, indicating independence between these items.

Understanding Lift in Association Rules

• To calculate lift for an association rule like milk and diapers → beer, one must divide the confidence by the support of beer. If calculated values show lift greater than one, it suggests increased likelihood due to knowledge about purchasing antecedents (milk and diapers).

• A lift value equal to one implies no effect on purchase probability; thus, knowing a customer bought milk does not influence their likelihood to buy beer at all. Conversely, a lift less than one indicates a negative impact on purchase likelihood when considering antecedents.

Implications of Lift Values

• If an association rule shows a lift less than one (e.g., milk → beer), it means customers who buy milk are less likely to buy beer compared to general buying patterns without this context. This highlights how certain products may negatively correlate with each other in consumer behavior analysis.