Claim Missing Document
Check
Articles

Found 2 Documents
Search

Customer Segmentation Based on Spending Patterns Using K-Means Clustering and PCA Dzakwan Akbar Perdana Wijaya; sasmita, Chesie fenta; Naufaldi Favian Archi
ROUTERS: Jurnal Sistem dan Teknologi Informasi Vol. 3 No. 2, Juli 2025 (In Progress)
Publisher : Program Studi Teknologi Rekayasa Internet, Politeknik Negeri Lampung

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.25181/rt.v3i2.3923

Abstract

Companies face challenges in understanding customer spending patterns, which can lead to ineffective marketing strategies. Traditional customer segmentation approaches often fail to accurately identify groups with different consumption behaviors. Therefore, this study proposes the implementation of the K-Means algorithm combined with Principal Component Analysis (PCA) to segment customers based on their spending patterns. This study uses a dataset containing customer spending information across various product categories, including wine, meat, fish, sweets, fruits, and gold. The Elbow method is applied to determine the optimal number of clusters, followed by K-Means clustering. The results are visualized using PCA to facilitate the interpretation of customer spending patterns. The findings indicate that the optimal number of clusters is six, with the Within-Cluster Sum of Squares (WCSS) decreasing from 50,000 for one cluster to 29,000 for six clusters. Cluster 3 exhibits the highest spending, particularly on meat at 566.91 and fish at 183.58, whereas Cluster 0 has the lowest spending, with its highest value being only 91.60 for wine. Silhouette Score evaluation shows that K-Means achieves a score of 0.4745, outperforming the Gaussian Mixture Model (GMM) with 0.0674
Design of an Edge-Intelligence Early-Warning System for Waste Odor Using Decision-Tree Classification-Rule Extraction on an ESP32 Microcontroller Agiska Ria Supriyatna; Imam Asrowardi; Eko Subyantoro; Dzakwan Akbar Perdana Wijaya; Naufaldi Favian Archi; Zahara Ummu Abdillah
Bulletin of Network Engineer and Informatics Vol. 4 No. 2 (2026): BUFNETS (Bulletin of Network Engineer and Informatics) October 2026
Publisher : PT. GWEX NET PUBLISHER

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.59688/fd5dmk21

Abstract

Uncontrolled organic waste releases volatile gases such as ammonia, hydrogen sulfide, and assorted volatile organic compounds that signal active decomposition and create both nuisance and health risk. Conventional odor monitoring relies on human olfaction or on laboratory instruments that are subjective, intermittent, or too costly for continuous deployment. This study presents a low-cost early-warning system that performs odor classification entirely at the edge on an ESP32 microcontroller. A gas sensor array (electronic nose) comprising six metal-oxide MQ sensors (MQ2, MQ3, MQ4, MQ8, MQ9, and MQ135) was used to acquire 184 labelled samples evenly split between a normal class and an odor-warning class. A Decision Tree classifier was trained, and the resulting decision logic was distilled into a compact set of human-readable threshold rules suitable for execution on a constrained microcontroller. Principal component analysis confirmed that the two odor states form clearly separable clusters (the first two components explain 89.7% of the variance). The extracted rule set attained 100% accuracy, precision, recall, and F1-score on a held-out test set, matching heavier models (random forest, support-vector machine, k-nearest neighbours) while requiring only a few bytes of memory and a handful of comparison operations per inference. To probe real-world reliability, a Gaussian sensor-noise injection study and a single-sensor-failure study were conducted. A redundant three-sensor majority-vote rule preserved above 95% accuracy at a noise level of 50% of the feature standard deviation and remained near 100% even when any single sensor failed, whereas a single-sensor rule collapsed to 48.6% on failure of its governing sensor. The results demonstrate that edge intelligence based on extracted decision-tree rules offers an accurate, interpretable, fault-tolerant, and extremely lightweight solution for continuous on-site waste-odor early warning.