<![CDATA[Transaction data at Selaras Coffee and Space had not been systematically utilized to evaluate menu combinations or determine which association rule mining algorithm best suited the data characteristics. This study compares Apriori, FP-Growth, and ECLAT for generating menu-bundling recommendations. Sales records from 1–31 December 2025 were preprocessed by removing 950 operational transaction-item pairs, resulting in 6,213 transactions and 76 unique menus. The algorithms were evaluated on the same binary matrix using a minimum support of 0.01, a minimum confidence of 0.20, and a lift ratio greater than 1. The evaluation included parameter sensitivity, 30 repeated measurements of execution time and peak Python memory allocation, scalability using 25–100% of the transactions, and rule quality based on support, confidence, lift, leverage, conviction, and cosine similarity. All algorithms produced identical outputs of 68 frequent itemsets and five eligible rules. On the full dataset, ECLAT recorded the lowest mean execution time at 0.037701 s, followed by Apriori at 0.040419 s and FP-Growth at 0.069637 s. FP-Growth used the lowest mean peak memory at 1.026966 MB, while ECLAT showed the lowest runtime growth as the dataset size increased. The strongest rule was Mie Laksa → Air Mineral 330 Ml, with a lift of 2.755987. These findings show that no algorithm dominated every criterion: ECLAT offered the best full-data runtime and scalability, FP-Growth was the most memory-efficient, and the extracted rules provided measurable candidates for menu-bundling strategies.]]>
Copyrights © 2026
