Data Mining Chapter 7: Association Rule Mining — Discovering Relationships

---

Opening Narrative: The Beer and Diapers Legend

In the early 1990s, a rumor began circulating through the corridors of data mining conferences that would become the field’s most enduring urban legend. According to the story, analysts at Walmart discovered an unexpected correlation in their transaction data: purchases of beer and diapers frequently occurred together, particularly on Thursday and Saturday evenings. The explanation offered was that young fathers, sent to purchase diapers for their newborns, would also pick up beer in anticipation of the weekend.

Whether the tale is apocryphal or grounded in actual analysis remains debated. Thomas Blischok, who led the early analytical efforts at Teradata working with retail data in the 1990s, has provided varying accounts over the years. Daniel Power’s research into the origin of this story suggests the correlation was discovered but the strategic placement of products was never implemented. Yet the legend persists because it captures something essential about association rule mining: the power to reveal non-obvious relationships hidden within vast transactional datasets.

What made this discovery conceptually revolutionary was not the statistical correlation itself—regression analysis could have found the same relationship—but the manner of discovery. Association rule mining does not require a hypothesis. It does not demand that an analyst ask “Is there a relationship between beer and diapers?” Instead, it systematically excavates all significant patterns from the data, surfacing relationships that no human analyst would think to investigate. This shift from hypothesis-driven to discovery-driven analysis represented a fundamental reconceptualization of how businesses could extract knowledge from their growing digital footprints.

The formal foundations for association rule mining were established in 1993 when Rakesh Agrawal, Tomasz Imielinski, and Arun Swami of IBM Almaden Research Center published their seminal paper introducing the problem of mining association rules in large databases. Their work, presented at the ACM SIGMOD conference, would accumulate over 35,000 citations and spawn an entire subfield of data mining research that continues to evolve three decades later.

---

Annotation

This chapter provides a comprehensive taxonomy of association rule mining methods, tracing the evolution from the foundational Apriori algorithm through modern constraint-based and multi-dimensional approaches. We examine the theoretical foundations of support, confidence, and lift metrics, present systematic classifications of algorithm families, and analyze real-world applications across retail, healthcare, and telecommunications. Five critical research gaps are identified, highlighting opportunities for addressing scalability limitations, incorporating temporal dynamics, and developing interpretable rule synthesis methods.

---

1. Introduction

Association rule mining represents one of the foundational paradigms in knowledge discovery, distinguished by its unsupervised, pattern-centric approach to revealing implicit relationships within transactional datasets. Unlike supervised learning methods that require labeled outcomes, or clustering algorithms that seek natural groupings, association rule mining operates on a fundamentally different principle: the exhaustive enumeration and evaluation of all co-occurrence patterns that satisfy user-specified thresholds.

The canonical form of an association rule is expressed as X → Y, where X (the antecedent) and Y (the consequent) are disjoint itemsets. The rule {bread, butter} → {milk} indicates that transactions containing bread and butter are likely to also contain milk. The strength of this relationship is quantified through three primary metrics:

• Support: The proportion of transactions containing both X and Y: sup(X∪Y) = |{t ∈ D : X∪Y ⊆ t}| / |D|
• Confidence: The conditional probability of Y given X: conf(X→Y) = sup(X∪Y) / sup(X)
• Lift: The ratio of observed to expected co-occurrence: lift(X→Y) = conf(X→Y) / sup(Y)

The challenge lies not in the mathematical simplicity of these metrics but in the computational complexity of evaluating them across exponentially large itemset spaces. A dataset with 1,000 unique items contains 2¹⁰⁰⁰ possible itemsets—a number exceeding the estimated atoms in the observable universe. Association rule mining algorithms must therefore employ sophisticated pruning strategies to render this search tractable.

graph TD
    A[Transaction Database] --> B[Frequent Itemset Mining]
    B --> C[Candidate Generation]
    C --> D[Support Counting]
    D --> E{Support ≥ min_sup?}
    E -->|Yes| F[Frequent Itemsets]
    E -->|No| G[Pruned]
    F --> H[Rule Generation]
    H --> I[Confidence Filtering]
    I --> J[Association Rules]
    
    style A fill:#e1f5fe
    style F fill:#c8e6c9
    style J fill:#fff9c4

Figure 1: The Association Rule Mining Pipeline

---

2. Problem Statement

Despite three decades of algorithmic development, association rule mining confronts several persistent challenges that limit its practical applicability:

The Pattern Explosion Problem: Lowering support thresholds to capture interesting but infrequent patterns results in combinatorial explosion. Real-world retail datasets with millions of transactions and tens of thousands of items can generate billions of candidate itemsets, overwhelming computational resources.

The Interestingness Challenge: High-support, high-confidence rules are not necessarily interesting or actionable. The rule {bread} → {milk} may have 95% confidence but reveals nothing novel. Distinguishing statistically significant patterns from actionable business insights remains fundamentally difficult.

The Temporal Blindness: Classical association rule mining treats transactions as independent, atemporal events. Yet purchasing patterns evolve seasonally, promotional campaigns create temporary correlations, and consumer preferences shift over time. Static rule sets become obsolete.

The Interpretability Burden: Industrial-scale mining operations can generate millions of rules. Without effective summarization, visualization, and prioritization mechanisms, decision-makers cannot extract actionable insights from this deluge.

---

3. Literature Review

The intellectual foundations of association rule mining emerge from the intersection of database theory, statistics, and machine learning. We trace the key contributions that shaped the field:

Foundational Work (1993-1995): Agrawal, Imielinski, and Swami’s 1993 paper established the formal problem definition. Their follow-up work with Srikant in 1994 introduced the Apriori algorithm, whose downward closure property—stating that all subsets of a frequent itemset must themselves be frequent—enabled efficient pruning of the search space. This paper alone has accumulated over 40,000 citations.

Algorithmic Advances (1995-2000): Han, Pei, and Yin’s 2000 FP-Growth algorithm represented a paradigm shift, eliminating candidate generation entirely through the use of compressed frequent pattern trees. The ECLAT algorithm by Zaki (2000) introduced vertical data representation, enabling efficient intersection-based support counting.

Interestingness Measures (1995-2010): Recognition that support and confidence were insufficient led to extensive research on alternative measures. Brin et al. (1997) introduced lift and conviction. Tan et al. (2002) provided a systematic comparison of 21 interestingness measures, demonstrating that no single measure dominates across all contexts.

Constraint-Based Mining (1998-2005): Ng et al. (1998) and Pei et al. (2001) developed frameworks for incorporating user-specified constraints—monotonic, anti-monotonic, and succinct—enabling more focused pattern discovery. This work addressed the pattern explosion problem by pushing constraints deep into the mining process.

Year	Contribution	Authors	Impact	Citations
1993	Problem Definition	Agrawal et al.	Foundational	35,000+
1994	Apriori Algorithm	Agrawal & Srikant	First Scalable Solution	40,000+
1997	Lift and Conviction	Brin et al.	Interestingness Measures	4,500+
2000	FP-Growth	Han, Pei, Yin	No Candidate Generation	25,000+
2000	ECLAT	Zaki	Vertical Mining	6,000+
2001	Constraint-Based Mining	Pei et al.	Focused Discovery	2,500+

Table 1: Foundational Contributions to Association Rule Mining

---

4. Taxonomic Classification of Association Rule Mining Methods

We present a comprehensive taxonomy organized along five orthogonal dimensions: algorithmic strategy, data representation, constraint handling, temporal modeling, and dimensionality. This multi-dimensional classification enables precise positioning of any association rule mining method within the broader methodological landscape.

graph LR
    A[Association Rule Mining Taxonomy] --> B[By Strategy]
    A --> C[By Representation]
    A --> D[By Constraints]
    A --> E[By Temporality]
    A --> F[By Dimensionality]
    
    B --> B1[Breadth-First: Apriori Family]
    B --> B2[Depth-First: ECLAT, FP-Growth]
    B --> B3[Hybrid: Partition, DHP]
    
    C --> C1[Horizontal: TID-Itemset]
    C --> C2[Vertical: Item-TIDset]
    C --> C3[Compressed: FP-Tree]
    
    D --> D1[Unconstrained]
    D --> D2[Item Constraints]
    D --> D3[Aggregate Constraints]
    D --> D4[Multi-Level]
    
    E --> E1[Static]
    E --> E2[Sequential]
    E --> E3[Temporal]
    E --> E4[Streaming]
    
    F --> F1[Single-Dimensional]
    F --> F2[Multi-Dimensional]
    F --> F3[Quantitative]
    
    style A fill:#e8eaf6
    style B fill:#c5cae9
    style C fill:#c5cae9
    style D fill:#c5cae9
    style E fill:#c5cae9
    style F fill:#c5cae9

Figure 2: Multi-Dimensional Taxonomy of Association Rule Mining

4.1 Algorithmic Strategy Classification

Breadth-First Algorithms (Apriori Family): These algorithms explore the itemset lattice level by level, generating all candidates of size k before moving to size k+1. The Apriori algorithm exploits the anti-monotonicity of support: if an itemset is infrequent, all its supersets must also be infrequent. Variants include AprioriTID (transaction encoding), AprioriHybrid (adaptive switching), and DHP (Direct Hashing and Pruning).

Depth-First Algorithms (ECLAT/FP-Growth Family): These traverse the itemset lattice depth-first, enabling memory-efficient mining of long patterns. ECLAT uses vertical data representation where each item maps to its transaction ID list (tidset). FP-Growth constructs a compressed frequent pattern tree and mines it recursively without candidate generation.

Hybrid Algorithms: Partition algorithms divide the database horizontally, mine local frequent itemsets in each partition, then validate globally. This approach reduces I/O costs and enables parallelization.

4.2 Data Representation Classification

graph TD
    subgraph Horizontal["Horizontal (TID-Itemset)"]
        T1[T1: Bread, Milk, Eggs]
        T2[T2: Bread, Butter]
        T3[T3: Milk, Eggs, Cheese]
    end
    
    subgraph Vertical["Vertical (Item-TIDset)"]
        V1["Bread: {T1, T2}"]
        V2["Milk: {T1, T3}"]
        V3["Eggs: {T1, T3}"]
    end
    
    subgraph Compressed["Compressed (FP-Tree)"]
        FP[Root] --> N1[Bread:2]
        N1 --> N2[Milk:1]
        N2 --> N3[Eggs:1]
        FP --> N4[Milk:1]
        N4 --> N5[Eggs:1]
    end
    
    style T1 fill:#e3f2fd
    style T2 fill:#e3f2fd
    style T3 fill:#e3f2fd
    style V1 fill:#f3e5f5
    style V2 fill:#f3e5f5
    style V3 fill:#f3e5f5
    style FP fill:#e8f5e9

Figure 3: Data Representation Paradigms for Association Rule Mining

4.3 Constraint Classification

Constraint-based mining addresses the pattern explosion by incorporating user knowledge into the mining process. Constraints are classified by their mathematical properties:

• Anti-monotonic: If a constraint is violated by itemset S, it is violated by all supersets of S. Example: support ≥ min_sup
• Monotonic: If satisfied by S, satisfied by all supersets. Example: sum(price) ≥ threshold
• Succinct: Can enumerate all satisfying itemsets directly. Example: items must include “bread”
• Convertible: Can be made anti-monotonic or monotonic through item reordering. Example: avg(price) ≥ threshold

Constraint Type	Example	Pruning Strategy	Complexity Impact
Anti-monotonic	support(X) ≥ 0.01	Prune supersets	Exponential reduction
Monotonic	sum(profit) ≥ $100	Prune subsets	Linear reduction
Succinct	X contains “luxury”	Enumerate directly	Domain restriction
Convertible	avg(price) ≥ $50	Order-dependent	Moderate reduction

Table 2: Constraint Classification and Pruning Strategies

4.4 Temporal and Sequential Extensions

Classical association rules ignore temporal ordering. Sequential pattern mining extends the framework to ordered sequences of itemsets. Key algorithmic developments include:

• GSP (Generalized Sequential Patterns): Apriori-style level-wise mining of sequences
• SPADE: Vertical sequence mining with efficient join operations
• PrefixSpan: Prefix-projection-based pattern growth without candidate generation
• Temporal Association Rules: Incorporate time windows, seasonal patterns, and decay functions

---

5. Algorithm Deep Dive

5.1 The Apriori Algorithm

Apriori remains the conceptual foundation for association rule mining education and serves as a baseline for performance comparison. The algorithm operates in two phases:

Phase 1 – Frequent Itemset Generation:

1. Scan database to count support of all individual items
2. Generate L₁: items meeting minimum support
3. For k = 2, 3, … until L_k-1 is empty:
   • Generate candidates C_k by joining L_k-1 with itself
   • Prune candidates with infrequent subsets
   • Scan database to count support of candidates
   • L_k = candidates meeting minimum support

Phase 2 – Rule Generation: For each frequent itemset L, generate all non-empty subsets S. For each subset, output rule S → (L – S) if confidence ≥ minimum confidence.

5.2 FP-Growth Algorithm

FP-Growth addresses Apriori’s key weakness: repeated database scans. The algorithm constructs a compressed representation of the database (FP-Tree) in two scans and then mines patterns directly from the tree:

1. First scan: Count item frequencies, sort items by frequency
2. Second scan: Build FP-Tree by inserting transactions as paths
3. Mining: For each item (bottom-up by frequency):
   • Extract conditional pattern base (prefix paths)
   • Construct conditional FP-Tree
   • Recursively mine conditional tree

FP-Growth achieves orders of magnitude speedup over Apriori for dense datasets with long patterns, as it requires only two database scans regardless of pattern length.

graph TD
    subgraph "FP-Tree Construction"
        R[Root] -->|2| A[A:2]
        A -->|1| B1[B:1]
        A -->|1| C1[C:1]
        B1 -->|1| C2[C:1]
        R -->|1| B2[B:1]
        B2 -->|1| C3[C:1]
    end
    
    subgraph "Conditional Pattern Mining"
        CP[Pattern: C] --> CPB["Conditional Base: {A,B:1}, {A:1}, {B:1}"]
        CPB --> CT[Conditional Tree]
        CT --> FP["Frequent Patterns: {A,C}, {B,C}, {A,B,C}"]
    end
    
    style R fill:#e8f5e9
    style A fill:#c8e6c9
    style B1 fill:#c8e6c9
    style B2 fill:#c8e6c9
    style C1 fill:#a5d6a7
    style C2 fill:#a5d6a7
    style C3 fill:#a5d6a7

Figure 4: FP-Tree Structure and Conditional Pattern Mining

5.3 Performance Comparison

Algorithm	DB Scans	Memory	Best For	Complexity
Apriori	k+1	O(candidates)	Sparse, short patterns	O(2|I|) worst case
FP-Growth	2	O(FP-Tree)	Dense, long patterns	O(|D| × |FP-Tree|)
ECLAT	1	O(tidsets)	Moderate density	O(tidset intersections)
LCM	1	O(|D|)	All pattern lengths	Linear per pattern

Table 3: Algorithm Performance Characteristics

---

6. Case Studies

This topic has been extensively analyzed by Oleh Ivchenko (February 2026) in Federated Learning for Privacy-Preserving Medical AI Training on the Stabilarity Research Hub, demonstrating how association patterns can be mined across distributed medical databases without centralizing sensitive patient data.

---

7. Multi-Dimensional and Quantitative Extensions

Classical association rules operate on boolean attributes: an item is either present or absent. Real-world applications require extensions to handle:

7.1 Multi-Dimensional Association Rules

Rules involving multiple dimensions or attributes. Example: age(X, “30-39”) ∧ income(X, “high”) → buys(X, “sports car”). Mining approaches include:

• Star-Schema Mining: Exploiting data warehouse star schemas for efficient multi-dimensional rule discovery
• Meta-Rule Guided Mining: User-specified templates constraining rule dimensions

7.2 Quantitative Association Rules

Rules involving numeric attributes. The challenge is determining appropriate discretization boundaries. Approaches include:

• Static Discretization: Pre-defined bins (age groups, income brackets)
• Dynamic Discretization: Srikant and Agrawal’s (1996) approach to merge adjacent intervals while maintaining minimum support
• Gradient-Based Methods: Optimize boundaries to maximize rule interestingness

graph LR
    subgraph "Boolean Rules"
        BR["X contains {bread, milk} → X contains {eggs}"]
    end
    
    subgraph "Multi-Dimensional Rules"
        MD["age='30-39' ∧ region='urban' → product='SUV'"]
    end
    
    subgraph "Quantitative Rules"
        QR["income ∈ [50K,80K] ∧ age ∈ [25,35] → credit_limit ∈ [10K,20K]"]
    end
    
    subgraph "Temporal Rules"
        TR["buy(X,laptop) →3mo buy(X,accessories)"]
    end
    
    style BR fill:#e3f2fd
    style MD fill:#f3e5f5
    style QR fill:#fff3e0
    style TR fill:#e8f5e9

Figure 5: Evolution of Association Rule Types

---

8. Interestingness Measures Beyond Confidence

The proliferation of interestingness measures reflects dissatisfaction with support and confidence as sole arbiters of rule quality. We present a taxonomic organization of 15+ measures:

Measure Categories:

• Objective Measures: Computed directly from data
  • Support, Confidence, Lift
  • Conviction: (1 – sup(Y)) / (1 – conf(X→Y))
  • Leverage: sup(X∪Y) – sup(X)×sup(Y)
  • Added Value: conf(X→Y) – sup(Y)
• Statistical Measures: Based on hypothesis testing
  • Chi-square
  • All-confidence: min(conf(A→B), conf(B→A))
  • Cosine: sup(X∪Y) / √(sup(X)×sup(Y))
• Information-Theoretic Measures:
  • J-measure: P(Y|X)×log(P(Y|X)/P(Y)) + P(¬Y|X)×log(P(¬Y|X)/P(¬Y))
  • Mutual Information

The comprehensive analysis of supervised learning quality measures by Oleh Ivchenko (February 2026) in Supervised Learning Taxonomy — Classification and Regression provides complementary perspectives on evaluating model quality that parallel interestingness measures in association rule mining.

---

9. Modern Applications and Platforms

Association rule mining has been implemented across numerous platforms and adapted for contemporary big data environments:

9.1 Apache Spark MLlib

Spark’s FP-Growth implementation enables distributed association rule mining across clusters. The parallel FP-Growth algorithm partitions the database, mines local frequent patterns, and aggregates results. Benchmarks demonstrate near-linear scalability to terabyte-scale datasets.

9.2 Python Ecosystem

• mlxtend: Provides Apriori and FP-Growth implementations with pandas integration
• Efficient-Apriori: Optimized implementation for large datasets
• Orange3-Associate: GUI-based association mining

The integration of data mining methods into modern ML pipelines has been documented by Oleh Ivchenko (February 2026) in Data Requirements and Quality Standards for Medical ML, demonstrating how association rules complement supervised learning in clinical decision support systems.

9.3 Enterprise Solutions

• SAS Enterprise Miner: Comprehensive association analysis with visualization
• IBM SPSS Modeler: Apriori and Carma algorithms
• KNIME: Open-source workflow-based association mining

---

10. Identified Gaps and Research Opportunities

Our systematic analysis of association rule mining literature reveals five critical gaps warranting further investigation:

The parallel challenge of privacy-preserving methods in medical contexts has been explored by Oleh Ivchenko (February 2026) in Federated Learning for Privacy-Preserving Medical AI Training, which addresses similar tensions between data utility and privacy protection.

graph TD
    A[Association Rule Mining Gap Landscape]
    A --> B[A7.1: Streaming at Scale]
    A --> C[A7.2: Causal Discovery]
    A --> D[A7.3: Cross-Domain Transfer]
    A --> E[A7.4: Ontological Mining]
    A --> F[A7.5: Privacy-Preserving]
    
    B --> B1[GPU Parallelism]
    B --> B2[Approximate Counting]
    B --> B3[Drift Detection]
    
    C --> C1[Do-Calculus Integration]
    C --> C2[Intervention Design]
    
    D --> D1[Domain Similarity]
    D --> D2[Rule Adaptation]
    
    E --> E1[Knowledge Graph Integration]
    E --> E2[Abstraction Selection]
    
    F --> F1[Differential Privacy]
    F --> F2[Secure Computation]
    
    style A fill:#ffebee
    style B fill:#ef9a9a
    style C fill:#ef9a9a
    style D fill:#ffe0b2
    style E fill:#fff9c4
    style F fill:#ffe0b2

Figure 6: Research Gap Taxonomy for Association Rule Mining

---

11. Conclusions

Association rule mining has evolved from a novel database technique to a foundational paradigm in knowledge discovery. The journey from Agrawal’s original Apriori algorithm to modern distributed implementations on platforms like Apache Spark represents three decades of sustained algorithmic innovation, driven by the relentless growth of transactional data.

Our taxonomic analysis reveals the field’s maturity along multiple dimensions: algorithmic strategies from breadth-first to depth-first approaches, data representations from horizontal to compressed tree structures, constraint frameworks from anti-monotonic to convertible constraints, and temporal extensions from static to streaming patterns. This systematic classification provides researchers and practitioners with a comprehensive map for navigating the methodological landscape.

The case studies from retail, healthcare, and telecommunications demonstrate that association rule mining continues to deliver practical value. Amazon’s recommendation engine, Stanford’s drug interaction discovery, and Korean Telecom’s churn prediction illustrate the technique’s versatility across domains with fundamentally different data characteristics and business objectives.

Yet five critical gaps remain. The challenge of streaming association mining at scale becomes increasingly urgent as IoT deployments generate petabytes of real-time data. The correlation-to-causation gap limits the prescriptive value of discovered patterns. Cross-domain transfer remains unexplored, forcing redundant mining efforts. Ontological integration promises richer, more interpretable patterns. And privacy-preserving methods must achieve practical utility-privacy tradeoffs to enable association mining in regulated domains.

The future of association rule mining likely lies at the intersection of these gaps: streaming causal discovery systems that operate on privacy-preserving federated data while leveraging ontological knowledge for interpretable pattern synthesis. Such systems would represent a qualitative leap beyond the current state of the art—and constitute a compelling research agenda for the decade ahead.

---

References

1. Agrawal, R., Imielinski, T., & Swami, A. (1993). Mining association rules between sets of items in large databases. Proceedings of the ACM SIGMOD International Conference on Management of Data, 207-216. https://doi.org/10.1145/170036.170072
2. Agrawal, R., & Srikant, R. (1994). Fast algorithms for mining association rules in large databases. Proceedings of the 20th International Conference on Very Large Data Bases, 487-499.
3. Han, J., Pei, J., & Yin, Y. (2000). Mining frequent patterns without candidate generation. ACM SIGMOD Record, 29(2), 1-12. https://doi.org/10.1145/335191.335372
4. Zaki, M. J. (2000). Scalable algorithms for association mining. IEEE Transactions on Knowledge and Data Engineering, 12(3), 372-390. https://doi.org/10.1109/69.846291
5. Brin, S., Motwani, R., Ullman, J. D., & Tsur, S. (1997). Dynamic itemset counting and implication rules for market basket data. ACM SIGMOD Record, 26(2), 255-264. https://doi.org/10.1145/253262.253325
6. Srikant, R., & Agrawal, R. (1996). Mining quantitative association rules in large relational tables. ACM SIGMOD Record, 25(2), 1-12. https://doi.org/10.1145/235968.233311
7. Tan, P. N., Kumar, V., & Srivastava, J. (2002). Selecting the right interestingness measure for association patterns. Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 32-41. https://doi.org/10.1145/775047.775053
8. Pei, J., Han, J., & Mao, R. (2000). CLOSET: An efficient algorithm for mining frequent closed itemsets. Proceedings of the ACM SIGMOD Workshop on Research Issues in Data Mining and Knowledge Discovery, 21-30.
9. Linden, G., Smith, B., & York, J. (2003). Amazon.com recommendations: Item-to-item collaborative filtering. IEEE Internet Computing, 7(1), 76-80. https://doi.org/10.1109/MIC.2003.1167344
10. Tatonetti, N. P., Ye, P. P., Daneshjou, R., & Altman, R. B. (2012). Data-driven prediction of drug effects and interactions. Science Translational Medicine, 4(125), 125ra31. https://doi.org/10.1126/scitranslmed.3003377
11. Kim, H. S., & Yoo, S. J. (2007). Enhanced prediction of customer churn in telecommunications. Expert Systems with Applications, 33(4), 870-880. https://doi.org/10.1016/j.eswa.2006.02.007
12. Han, J., & Fu, Y. (1995). Discovery of multiple-level association rules from large databases. Proceedings of the 21st International Conference on Very Large Data Bases, 420-431.
13. Ng, R. T., Lakshmanan, L. V., Han, J., & Pang, A. (1998). Exploratory mining and pruning optimizations of constrained associations rules. ACM SIGMOD Record, 27(2), 13-24. https://doi.org/10.1145/276305.276307
14. Pei, J., Han, J., & Lakshmanan, L. V. (2001). Mining frequent itemsets with convertible constraints. Proceedings of the 17th International Conference on Data Engineering, 433-442. https://doi.org/10.1109/ICDE.2001.914856
15. Kantarcioglu, M., & Clifton, C. (2004). Privacy-preserving distributed mining of association rules on horizontally partitioned data. IEEE Transactions on Knowledge and Data Engineering, 16(9), 1026-1037. https://doi.org/10.1109/TKDE.2004.45
16. Uno, T., Asai, T., Uchida, Y., & Arimura, H. (2004). An efficient algorithm for enumerating closed patterns in transaction databases. Proceedings of Discovery Science, 16-31. https://doi.org/10.1007/978-3-540-30214-8_2
17. Gionis, A., Mannila, H., Mielikäinen, T., & Tsaparas, P. (2007). Assessing data mining results via swap randomization. ACM Transactions on Knowledge Discovery from Data, 1(3), Article 14. https://doi.org/10.1145/1297332.1297338
18. Agrawal, R., & Srikant, R. (1995). Mining sequential patterns. Proceedings of the 11th International Conference on Data Engineering, 3-14. https://doi.org/10.1109/ICDE.1995.380415
19. Pei, J., Han, J., Mortazavi-Asl, B., & Pinto, H. (2001). PrefixSpan: Mining sequential patterns efficiently by prefix-projected pattern growth. Proceedings of the 17th International Conference on Data Engineering, 215-224.
20. Zaki, M. J. (2001). SPADE: An efficient algorithm for mining frequent sequences. Machine Learning, 42(1), 31-60. https://doi.org/10.1023/A:1007652502315
21. Park, J. S., Chen, M. S., & Yu, P. S. (1997). Using a hash-based method with transaction trimming for mining association rules. IEEE Transactions on Knowledge and Data Engineering, 9(5), 813-825. https://doi.org/10.1109/69.634757
22. Savasere, A., Omiecinski, E., & Navathe, S. B. (1995). An efficient algorithm for mining association rules in large databases. Proceedings of the 21st International Conference on Very Large Data Bases, 432-444.
23. Geng, L., & Hamilton, H. J. (2006). Interestingness measures for data mining: A survey. ACM Computing Surveys, 38(3), Article 9. https://doi.org/10.1145/1132960.1132963
24. Power, D. J. (2002). What is the “true story” about data mining, beer and diapers? DSS News, 3(23). Retrieved from https://www.dssresources.com/newsletters/66.php
25. Verma, K., Singh, S., & Sethia, D. (2020). A comprehensive survey on frequent pattern mining algorithms for graph data. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 10(6), e1391. https://doi.org/10.1002/widm.1391

---

Next Chapter: Chapter 8 — Sequential Pattern Mining: Temporal Discoveries