<![CDATA[This study analyzes the performance of the Frequent Pattern Growth (FP-Growth) algorithm on motorcycle service data based on transmission-type and service-category segmentation at AHASS Roda Motor Cianjur. The dataset consists of 398 observation rows and 13 service items obtained from monthly Dealer Management System records covering August 2024 to February 2026. Following the Knowledge Discovery in Databases framework, the data were cleaned, selected, segmented, and transformed into a binary matrix. A value of 1 was assigned when the service volume exceeded 10, while a value of 0 was assigned when the volume was 10 or less. Five testing scenarios were constructed: combined data, automatic-transmission motorcycles, manual-clutch motorcycles, free services, and paid services. The algorithm was executed using a minimum support of 0.30, a minimum confidence of 0.70, and a lift ratio greater than one. Performance evaluation was conducted in Visual Studio Code using three warm-up runs and 30 measurement repetitions to record execution time and the increase in Resident Set Size. The results show that the combined dataset produced 62 frequent itemsets and 154 association rules. The automatic-transmission motorcycle scenario produced 304 frequent itemsets and 4,642 association rules, while the manual-clutch motorcycle scenario produced seven frequent itemsets and two rules. The free-service scenario generated seven frequent itemsets and 11 rules, whereas the paid-service scenario generated 29 frequent itemsets and 66 rules. The combined dataset recorded a median total execution time of 8.7783 milliseconds and a median memory increase of 0.4141 MB. The automatic-transmission motorcycle scenario had the highest computational load, requiring 44.3190 milliseconds and 3.9023 MB, whereas the manual-clutch motorcycle scenario recorded the lowest execution time of 5.4048 milliseconds and a memory increase of 0.3945 MB. The free-service and paid-service scenarios required 6.1301 milliseconds and 7.2973 milliseconds, respectively. These findings indicate that segmentation produces more specific association patterns but does not always improve computational efficiency. FP-Growth performance is influenced by data density, active item count, frequent itemset count, and association-rule complexity. Therefore, fewer observation rows do not necessarily result in lower computational requirements.]]>
