Pemodelan Prediksi Pelanggan Potensial Untuk Tambah Daya Menggunakan Gradient Boosting Berdasarkan Pola Konsumsi Listrik (Studi Kasus PLN UP3 GPI)

Sumpena, Asep (2026) Pemodelan Prediksi Pelanggan Potensial Untuk Tambah Daya Menggunakan Gradient Boosting Berdasarkan Pola Konsumsi Listrik (Studi Kasus PLN UP3 GPI). Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6032241040-Master_Thesisi.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

PT PLN (Persero) UP3 GPI menghadapi tantangan dalam meningkatkan penjualan listrik, terutama di segmen pelanggan tegangan menengah dan tinggi yang meskipun jumlahnya sedikit, namun berkontribusi besar terhadap pendapatan perusahaan. Salah satu strategi utama dalam peningkatan penjualan listrik adalah program tambah daya. Namun, proses identifikasi pelanggan potensial selama ini masih dilakukan secara konvensional sehingga kurang efisien dan sering tidak tepat sasaran. Oleh karena itu, penelitian ini penting untuk mengembangkan model rekomendasi berbasis supervised machine learning guna mengidentifikasi pelanggan potensial mana yang menjadi target tambah daya secara lebih akurat dan proaktif. Penelitian ini menggunakan data internal PLN UP3 GPI tahun 2023–2025 yang mencakup 467 pelanggan TM/TT dengan 16 variabel, meliputi karakteristik pelanggan, profil langganan, klasifikasi tarif, dan pola konsumsi listrik. Variabel respon adalah status tambah daya (YA/TIDAK) dengan rasio ketidakseimbangan kelas 7,65:1 (88,44% TIDAK, 11,56% YA). Proses penelitian meliputi eksplorasi data, pra-pemrosesan, pemodelan prediksi menggunakan algoritma Gradient Boosting, serta evaluasi model dengan metrik akurasi, precision, recall, F1-score, AUC, MCC, dan specificity. Tiga skenario pembagian data dievaluasi menggunakan stratified random sampling: 70:30, 80:20, dan 90:10. Skenario 70:30 menghasilkan performa paling andal: akurasi 91,4%, precision 66,7%, recall 50,0%, F1-score 57,1%, AUC 0,722, MCC 0,532, dan specificity 96,8% (TN=120, FP=4, FN=8, TP=8) pada n=140 data uji. Skenario 80:20 menghasilkan AUC 0,590 dan recall 27,3%, sedangkan skenario 90:10 mencapai AUC 0,751 namun hanya dengan 5 instansi YA pada data uji, sehingga keandalannya terbatas. Model memungkinkan PLN menargetkan 8 dari setiap 12 kandidat upgrade secara tepat, peningkatan 6–8 kali dibanding tingkat konversi konvensional 5–10%. Model dapat diimplementasikan untuk mendukung strategi targeted marketing tambah daya, perencanaan infrastruktur, segmentasi pelanggan, serta monitoring melalui dashboard real-time, sehingga mendukung transformasi digital PLN UP3 GPI secara berkelanjutan.
=======================================================================================================================================
PT PLN (Persero) UP3 GPI faces challenges in increasing electricity sales, particularly among medium and high voltage customers who, despite their small numbers, make a significant contribution to company revenue. One of the main strategies for boosting electricity sales is the power capacity upgrade program. However, the process of identifying potential customers for upgrades has so far been conducted conventionally, resulting in inefficiency and often inaccurate targeting. Therefore, this study is important for developing a recommendation model based on supervised machine learning to identify potential customers for power upgrades more accurately and proactively. This research uses internal PLN UP3 GPI data from 2023–2025, comprising 467 TM/TT customers with 16 variables covering customer characteristics, subscription profiles, tariff classifications, and electricity consumption patterns. The response variable is the power upgrade status (YES/NO) with a 7.65:1 class imbalance ratio (88.44% NO, 11.56% YES). The research process includes data exploration, preprocessing, predictive modeling using the Gradient Boosting algorithm, and model evaluation using accuracy, precision, recall, F1-score, AUC, MCC, and specificity metrics. Three data-splitting scenarios were evaluated using stratified random sampling: 70:30, 80:20, and 90:10. The 70:30 split yielded the most reliable performance: accuracy 91.4%, precision 66.7%, recall 50.0%, F1-score 57.1%, AUC 0.722, MCC 0.532, and specificity 96.8% (TN=120, FP=4, FN=8, TP=8) on n=140 test samples. The 80:20 split yielded AUC 0.590 and recall 27.3%, while the 90:10 split achieved AUC 0.751 but with only 5 YES test instances, limiting its reliability. The model enables PLN to correctly target 8 of every 12 predicted upgrade candidates, a 6–8× improvement over conventional 5–10% conversion rates. It can be deployed to support targeted marketing, infrastructure planning, customer segmentation, and real-time dashboard monitoring, thereby advancing PLN UP3 GPI’s digital transformation.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	gradient boosting, konsumsi energi, machine learning, pelanggan potensial, tambah daya
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T58.6 Management information systems
Divisions:	Interdisciplinary School of Management and Technology (SIMT) > 61101-Master of Technology Management (MMT)
Depositing User:	Asep Sumpena
Date Deposited:	30 Jul 2026 04:16
Last Modified:	30 Jul 2026 04:16
URI:	http://repository.its.ac.id/id/eprint/141251

Actions (login required)

View Item