<![CDATA[high-volume products do not necessarily yield proportionate profit contributions if their cost of goods sold is also high. This study aims to develop a product portfolio segmentation approach that integrates profitability dimensions into clustering and association rule mining processes. The dataset consists of 13,159 transactions and 22,185 itemized rows from Jalan Cerita Kopi & Space over a 180-day period from January to June 2026, supplemented by cost of goods sold data for all 59 products. The methodology employs K-Means clustering utilizing six derived features capturing sales volume, profit margin, total profit contribution, and purchasing behavior. Validation is performed via Elbow, Silhouette Coefficient, Davies-Bouldin Index, and Calinski-Harabasz Index, and validated against Agglomerative Hierarchical Clustering. Association rule mining is executed using the FP-Growth algorithm, after which each rule is re-evaluated using a profit-based utility metric. The results yield four distinct product clusters with a Silhouette score of 0.457, a Davies-Bouldin Index of 0.873, and an agreement level of 0.883 with Agglomerative Clustering measured by Adjusted Rand Index. The primary finding reveals that association rule rankings based on lift and profit utility are virtually uncorrelated, with a Spearman correlation coefficient of only 0.168 and zero overlap among the top ten rules. The rule with the highest lift generated Rp 1,367,000 in profit, whereas the rule with the highest utility yielded Rp 3,309,000 despite a modest lift of 1.08. These findings demonstrate that bundling strategies designed solely on frequency metrics risk guiding business owners toward product combinations that are frequently purchased yet financially sub-optimal.]]>
Copyrights © 2026
