Frequent itemset mining is a key task in data mining, particularly for analyzing customer purchase behavior in large-scale transaction datasets. This paper proposes a novel method called SPGAN-ECG-FISM-LSDA—Semantic Preserved Generative Adversarial Network-driven Enhanced Candidate Generation for Efficient and Scalable Frequent Itemset Mining. The approach begins by converting transaction data into bit vectors and generating compact candidate itemsets using subset discovery, thereby reducing memory usage and database scans. To model and predict customer purchasing behavior, a semantic-preserved GAN (SPGAN) is applied, and its parameters are optimized using the Shrike Optimization Algorithm (SOA) for enhanced accuracy and efficiency. The proposed approach is implemented in Python and evaluated against benchmark models such as EM-THUI-GA, FIFS-RMSA-DC, and EM-CLHUI-TQD. Experimental results demonstrate significant improvements in scanning time, processing time, and storage efficiency, confirming the effectiveness of the SPGAN-ECG-FISM-LSDA framework for large-scale data analytics.
Rajalakshmi et al. (Mon,) studied this question.