Market Basket / Association Analysis

business analytics

slug: topic-map-business-analytics-market-basket-association-analysis

Vocabulary:

• market_basket: Set of items purchased together
• association_rule: If X then Y relationship (X→Y)
• antecedent: The "if" part (left-hand side) of rule
• consequent: The "then" part (right-hand side) of rule
• support: Frequency of itemset (P(X))
• confidence: Conditional probability (P(Y|X))
• lift: Ratio of observed to expected co-occurrence (P(Y|X) / P(Y))
• conviction: Measure of implication strength
• leverage: Difference between observed and expected co-occurrence
• itemset: Collection of items
• frequent_itemset: Itemset appearing above minimum support threshold
• apriori: Classic algorithm for finding association rules

Concepts:

• co_occurrence: Items appearing together more than by chance
• positive_association: Lift > 1, items appear together more than expected
• negative_association: Lift < 1, items appear together less than expected
• independence: Lift ≈ 1, no association
• rule_quality: Balancing support, confidence, and lift
• actionable_rules: Rules useful for business decisions
• spurious_association: Statistical association without causal relationship
• minimum_thresholds: Support, confidence, lift cutoffs to filter rules
• rule_redundancy: Multiple rules saying essentially same thing

Concepts_advanced:

• closed_itemsets: Itemsets with no superset of same support
• maximal_itemsets: Itemsets with no superset above minimum support
• sequential_patterns: Order matters (A then B then C)
• hierarchical_associations: Rules at different aggregation levels
• multi_dimensional_associations: Rules involving multiple attributes beyond items
• negative_rules: X→NOT Y relationships

Procedures:

• identify_transactions: Define what constitutes a "basket"
• create_itemsets: List all unique items or dimension values
• calculate_support: Count frequency of each itemset
• filter_by_min_support: Keep only frequent itemsets
• generate_rules: Create X→Y from frequent itemsets
• calculate_confidence: P(Y|X) for each rule
• calculate_lift: Observed/expected ratio
• filter_by_min_confidence: Keep strong rules
• filter_by_min_lift: Keep rules with lift > 1 (or higher threshold)
• rank_rules: Order by lift, confidence, support, or combination

Procedures_detailed:

• support_calculation: count(X) / total_transactions
• confidence_calculation: count(X,Y) / count(X)
• lift_calculation: confidence(X→Y) / support(Y) = P(Y|X) / P(Y)
• conviction_calculation: [1 - support(Y)] / [1 - confidence(X→Y)]
• leverage_calculation: support(X,Y) - [support(X) × support(Y)]
• apriori_algorithm: Iteratively generate candidate itemsets, prune infrequent

Topics:

• product_recommendations
• cross_sell_optimization
• bundle_design
• store_layout_optimization
• customer_segment_profiling
• content_recommendation
• fraud_pattern_detection
• diagnostic_associations
• clickstream_analysis
• menu_optimization

Categories:

• pattern_mining
• co_occurrence_analysis
• rule_discovery
• recommendation_systems
• behavioral_analysis

Themes:

• discovering_hidden_patterns: Find non-obvious associations
• actionable_recommendations: Rules that drive business decisions
• statistical_vs_causal: Association doesn't imply causation
• incremental_revenue: Driving add-on purchases

Trends:

• personalized_associations: Rules specific to customer segments
• real_time_recommendations: Live association-based suggestions
• deep_learning_embeddings: Neural nets learn item associations
• graph_based_associations: Network analysis of item relationships
• contextual_associations: Rules vary by time, location, season

Use_cases:

• retail: "Customers who buy diapers also buy beer (lift=2.3) - place nearby or bundle"
• ecommerce: "Users viewing laptops then view laptop bags (confidence=65%, lift=3.1)"
• restaurant: "Customers ordering steak also order wine (lift=2.8) - train upsell"
• streaming: "Users watching Show A also watch Show B (lift=4.2) - recommend"
• healthcare: "Patients with Symptom X often have Diagnosis Y (lift=3.5) - screening protocol"
• banking: "Customers with checking also get savings (lift=2.1) - cross-sell campaign"
• telecom: "Subscribers with unlimited data also get streaming bundle (lift=3.9)"
• hospitality: "Guests booking spa also book late checkout (lift=2.6) - package deal"