Pengembangan Kerangka Analitis Segmentasi Multikriteria Berbasis K-Means Dan Peramalan Adaptif Untuk Suku Cadang Di PLTU XYZ

Saputro, Ferdian Harry (2026) Pengembangan Kerangka Analitis Segmentasi Multikriteria Berbasis K-Means Dan Peramalan Adaptif Untuk Suku Cadang Di PLTU XYZ. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6010241114-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Peningkatan kompleksitas operasi pembangkit listrik menjadikan pengelolaan persediaan suku cadang semakin krusial. Data dashboard inventaris PLTU XYZ menunjukkan bahwa nilai persediaan meningkat dari sekitar Rp11,85 miliar pada 2023 menjadi Rp20,79 miliar pada 2025, sementara inventory turnover menurun dari 4,23 menjadi 1,99 kali per tahun dan nilai deadstock masih berada pada kisaran Rp3–4 miliar. Kondisi ini mengindikasikan bahwa pengendalian persediaan belum optimal, terutama karena sistem eksisting masih bertumpu pada klasifikasi material dan evaluasi enjiniring, sementara peramalan permintaan belum digunakan secara sistematis sebagai dasar perencanaan kebutuhan suku cadang. Di sisi lain, kajian literatur dan analisis bibliometrik menunjukkan bahwa inventory control, machine learning–based clustering, dan demand forecasting untuk suku cadang masih banyak berkembang secara terpisah, sehingga diperlukan kerangka analitis yang menghubungkan segmentasi material dan peramalan adaptif.
Penelitian ini mengembangkan kerangka analitis berbasis K-Means clustering dan peramalan adaptif pada PLTU XYZ. Klasterisasi dilakukan terhadap 1.215 item menggunakan lima atribut, yaitu criticality, availability, usage value, ADI, dan CV², serta menghasilkan empat klaster optimal dengan mean silhouette coefficient sebesar 0,45. Klaster 1 merupakan High Value Operational and Maintenance Part dengan frekuensi pemakaian tinggi dan variabilitas permintaan tertinggi (CV² = 0,651). Klaster 2 merupakan Strategic Spare Part yang berisi item bernilai tinggi, memiliki permintaan sangat jarang (ADI ≥ 8,5), serta memiliki kriteria suku cadang paling kritikal terbanyak dari seluruh klaster. Klaster 3 merupakan General–Slow Moving Spare Part yang bernilai rendah dan jarang digunakan, sedangkan Klaster 4 merupakan Low Value–Routine and Consumable Part sebagai kelompok terbesar dengan permintaan relatif lebih reguler. Klasterisasi multikriteria terbukti mampu mengidentifikasi esensi item “strategis” pada Klaster 2 yang tidak seluruhnya tertangkap oleh sistem klasifikasi konvensional perusahaan yang hanya berbasis criticality dan nilai.
Pada tahap peramalan, sembilan item sampel dievaluasi menggunakan SES, Croston, SBA, TSB, mSBA, dan mTSB secara adaptif per item melalui grid search parameter. Hasil menunjukkan SBA terbaik pada empat item, mSBA pada tiga item, serta Croston dan TSB masing-masing pada satu item; sedangkan SES dan mTSB tidak terpilih sebagai metode terbaik. Temuan ini menegaskan bahwa kuadran ADI–CV² berguna untuk mengklasifikasikan karakteristik permintaan, tetapi tidak menjamin kesamaan metode peramalan optimal. Evaluasi rolling-origin pada tujuh skema pembagian data menunjukkan bahwa lima dari sembilan item sampel memiliki dominasi metode terbaik sebesar 85,7%–100%, yang mengindikasikan kestabilan hasil terhadap perubahan periode evaluasi. Kerangka ini dapat menjadi dasar decision support dalam penyusunan inventory policy, pemetaan persediaan, evaluasi suku cadang, dan pengembangan pengendalian persediaan yang lebih adaptif di PLTU XYZ.
====================================================================================================================================
The increasing complexity of power plant operations has made spare parts inventory management increasingly critical. Inventory dashboard data from PLTU XYZ indicate that inventory value increased from approximately IDR 11.85 billion in 2023 to IDR 20.79 billion in 2025, while inventory turnover declined from 4.23 to 1.99 times per year, and deadstock value remained within the range of IDR 3–4 billion. This condition indicates that inventory control has not yet been optimal, particularly because the existing system still relies on material classification and engineering evaluation, while demand forecasting has not been systematically used as a basis for spare parts requirement planning. In addition, the literature review and bibliometric analysis show that studies on inventory control, machine learning-based clustering, and demand forecasting for spare parts have largely developed separately. Therefore, an analytical framework that links material segmentation and adaptive forecasting is required.
This study develops an analytical framework based on K-Means clustering and adaptive forecasting at PLTU XYZ. Clustering was performed on 1,215 items using five attributes, namely criticality, availability, usage value, ADI, and CV², resulting in four optimal clusters with a mean silhouette coefficient of 0.45. Cluster 1 represents High Value Operational and Maintenance Part, characterized by high usage frequency and the highest demand variability (CV² = 0.651). Cluster 2 represents Strategic Spare Part, consisting of high-value items with very rare demand occurrences (ADI ≥ 8.5) and the highest concentration of highly critical spare parts among all clusters. Cluster 3 represents General–Slow Moving Spare Part, consisting of low-value and rarely used items, while Cluster 4 represents Low Value–Routine and Consumable Part as the largest group with relatively more regular demand. The multicriteria clustering approach demonstrates its ability to identify the strategic nature of items in Cluster 2, which is not fully captured by the company’s conventional classification system that relies mainly on criticality and value-based criteria.
In the forecasting stage, nine sample items were evaluated using SES, Croston, SBA, TSB, mSBA, and mTSB adaptively at the item level through parameter grid search. The results show that SBA was the best-performing method for four items, mSBA for three items, while Croston and TSB were each selected as the best method for one item. Meanwhile, SES and mTSB were not selected as the best-performing methods for any item. These findings confirm that the ADI–CV² quadrant is useful for classifying demand characteristics but does not guarantee the same optimal forecasting method. Rolling-origin evaluation across seven data-splitting schemes shows that five out of nine sample items achieved best-method dominance of 85.7%–100%, indicating the stability of the results against changes in the evaluation period. The proposed framework can serve as a decision-support basis for developing Inventory Policy, inventory mapping, spare parts evaluation, and a more adaptive spare parts inventory control system at PLTU XYZ.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	manajemen persediaan, k-means clustering, permintaan intermiten , peramalan adaptif, unsupervised learning,inventory management, k-means clustering, intermittent demand, adaptive forecasting, unsupervised learning.
Subjects:	Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) Q Science > QA Mathematics > QA76.9.D37 Data warehousing. Q Science > QA Mathematics > QA278 Cluster Analysis. Multivariate analysis. Correspondence analysis (Statistics)
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Industrial Engineering > 26101-(S2) Master Thesis
Depositing User:	Ferdian Harry Saputro
Date Deposited:	29 Jul 2026 06:27
Last Modified:	29 Jul 2026 06:27
URI:	http://repository.its.ac.id/id/eprint/140029

Actions (login required)

View Item