<![CDATA[Retail transaction data provide useful insights into the products customers tend to purchase together. However, identifying these purchasing patterns can be challenging when a dataset contains thousands of different products. This study investigates the effect of minimum support on the results and execution time of the Equivalence Class Transformation (ECLAT) algorithm for market basket analysis. The study also implements the analysis in a web-based application. The experiment uses the Online Retail dataset from the UCI Machine Learning Repository. After preprocessing, 530,104 records were retained, forming 19,960 transactions involving 3,812 unique products. Five minimum support thresholds were evaluated, ranging from 1% to 3%, while minimum confidence was set to 50% and the maximum itemset length was limited to six items. The results show that increasing minimum support from 1% to 3% reduced the number of frequent itemsets from 2,026 to 142 and the number of association rules from 994 to 12. Execution time also decreased from 303.2885 to 95.9162 seconds under the tested conditions. At the 3% minimum support threshold, all 12 generated rules had lift values greater than one. The strongest representative rule connected pink regency teacup and saucer with green regency teacup and saucer, achieving a confidence of 82.613% and a lift of 16.278. The web-based application brings together data preprocessing, frequent pattern mining, parameter testing, result visualization, and data export. Overall, the findings demonstrate how minimum support influences the number of patterns identified and the execution time observed in this experiment.]]>
Copyrights © 2026
