[Paper Review] An Algorithm for Mining High Utility Closed Itemsets and Generators

Jayakrushna Sahoo, Ashok Kumar Das+1 more|arXiv (Cornell University)|Oct 11, 2014

Data Mining Algorithms and Applications42 references3 citations

TL;DR

This paper proposes HUCI-Miner, a novel algorithm for mining high utility closed itemsets (HUCI) and their minimal generators simultaneously, enabling condensed, non-redundant association rule generation in high utility itemset mining. By extending the concept of minimal generators from traditional support-confidence mining to utility-based mining, the method achieves significant compression of high utility itemsets while preserving rule semantics.

ABSTRACT

Traditional association rule mining based on the support-confidence framework provides the objective measure of the rules that are of interest to users. However, it does not reflect the utility of the rules. To extract non-redundant association rules in support-confidence framework frequent closed itemsets and their generators play an important role. To extract non-redundant association rules among high utility itemsets, high utility closed itemsets (HUCI) and their generators should be extracted in order to apply traditional support-confidence framework. However, no efficient method exists at present for mining HUCIs with their generators. This paper addresses this issue. A post-processing algorithm, called the HUCI-Miner, is proposed to mine HUCIs with their generators. The proposed algorithm is implemented using both synthetic and real datasets.

Motivation & Objective

• To address the lack of efficient methods for mining high utility closed itemsets (HUCI) with their generators.
• To enable condensed, non-redundant association rule mining in high utility itemset mining by integrating minimal generator concepts.
• To reduce the size of high utility itemsets and associated rule sets while preserving semantic and statistical relevance.
• To provide a post-processing framework that supports traditional support-confidence rule generation on high utility patterns.
• To demonstrate the effectiveness of HUCI-Miner on both synthetic and real-world datasets with varying utility thresholds.

Proposed method

• Proposes a level-wise search algorithm, HUCI-Miner, to mine high utility itemsets (HUI), high utility closed itemsets (HUCI), and their minimal generators (HG) in a single pass.
• Uses utility unit arrays to reconstruct all HUIs from HUCIs, enabling efficient rule generation without full database scans.
• Introduces the concept of minimal generators in high utility mining, where a generator is a minimal itemset that uniquely generates a closed itemset.
• Applies a closure-based approach to identify HUCIs by checking if an itemset has the same utility as its supersets.
• Implements a post-processing step to extract generators from HUCIs by verifying minimality through subset utility comparison.
• Employs main-memory storage for high utility itemsets to accelerate processing and improve performance.

Experimental results

Research questions

• RQ1 How can high utility closed itemsets (HUCI) and their minimal generators be efficiently mined in a single algorithmic framework?
• RQ2 To what extent can HUCI-Miner compress the number of high utility itemsets compared to full enumeration?
• RQ3 Can the integration of minimal generators into high utility mining effectively reduce redundancy in extracted association rules?
• RQ4 How does the performance of HUCI-Miner scale across diverse real and synthetic datasets with varying utility thresholds?
• RQ5 What is the impact of different minimum utility thresholds on the number of HUCIs and generators discovered?

Key findings

• HUCI-Miner achieved up to 80% reduction in the number of high utility itemsets by compressing them into HUCIs and their generators.
• On the T10I4D100K dataset, at 0.006% min_util, HUCI-Miner reduced HUI count from 192,673 to 111,544 HUCIs and 28,837 generators, totaling 140,381 compressed patterns.
• For the dense Chess dataset at 25% min_util, HUCI-Miner reduced HUIs from 13,331 to 8,700 HUCIs and 3,811 generators, achieving a 34.5% reduction.
• In the Retail dataset, all HUIs were closed (100% closure rate), and only 2 generators were found at 0.02% min_util, indicating high compression efficiency.
• The Chain-store dataset showed 100% closure rate across all thresholds, with no non-self-generators found, indicating that all HUCIs were maximal.
• The algorithm demonstrated scalability on large datasets like Chain-store (1.1M transactions), with HUCI and HG counts growing predictably with increasing min_util thresholds.