Optimasi Pendeteksian Pencurian Tenaga Listrik Berbasis Automated Machine Learning Melalui Pendekatan Multivariabel dan Cost-Sensitive Learning: Studi Kasus di PT PLN (Persero) UID Jawa Barat

Arla, Dino (2025) Optimasi Pendeteksian Pencurian Tenaga Listrik Berbasis Automated Machine Learning Melalui Pendekatan Multivariabel dan Cost-Sensitive Learning: Studi Kasus di PT PLN (Persero) UID Jawa Barat. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6032231231-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (12MB) | Request a copy

Abstract

Pencurian tenaga listrik merupakan permasalahan serius yang berdampak signifikan pada kerugian ekonomi, efisiensi operasional, dan stabilitas keuangan perusahaan utilitas. PT PLN (Persero) Unit Induk Distribusi (UID) Jawa Barat sebagai unit distribusi terbesar di Indonesia juga menghadapi tantangan besar dalam menangani masalah ini. Pendekatan deteksi tradisional selama ini masih terbatas pada analisa pemakaian energi dengan asumsi subjektif, tanpa prioritas strategis dalam pemeriksaan langsung ke pelanggan (on site investigation). Penelitian ini bertujuan mengoptimalkan pendeteksian pencurian listrik di PLN UID Jawa Barat menggunakan Automated Machine Learning (AutoML) dengan pendekatan multivariabel dan Cost-Sensitive Learning. Variabel yang dipilih mencakup fitur perilaku pelanggan, data teknis jaringan, serta atribut geografis dan sosial ekonomi yang diolah dari lebih dari 6,4 juta data pemeriksaan pelanggan riil pada tahun 2019 hingga 2023. Sebanyak 14 algoritma klasifikasi, termasuk Gradient Boosting, Decision Tree, KNN, Logistic Regression hingga Random Forest, dievaluasi dengan metode stratified cross-validation. Model Extreme Gradient Boosting Classifier (XGBoost) menunjukkan performa terbaik dengan F1-score 0,92, AUC 0,87 pada data holdout serta mampu mendeteksi 93% kasus dengan precision sebesar 93%. Implementasi Cost-Sensitive Learning dengan penalti berbeda terhadap jenis kesalahan klasifikasi meningkatkan jumlah deteksi temuan valid (true positive) secara signifikan dari 22.883 menjadi 75.884 pelanggan setelah optimasi, tanpa menambah tingkat kesalahan. Model hasil optimasi ini diimplementasikan ke dalam sistem operasional PLN UID Jawa Barat dan terintegrasi dengan aplikasi internal EPM. Pipeline otomatis yang dikembangkan tidak hanya mempercepat analisis tetapi juga mendukung pengambilan keputusan berbasis data terkait prioritas investigasi di lapangan. Analisis cost-benefit menunjukkan potensi penghematan kerugian akibat pencurian listrik mencapai 334 miliar rupiah, dengan tingkat pengembalian investasi (ROI) sebesar 13,57%. Risk analysis dilakukan untuk memastikan keamanan implementasi model melalui validasi berkala, integrasi keamanan data, serta pelatihan pengguna. Keseluruhan hasil penelitian ini menegaskan efektivitas transformasi digital berbasis kecerdasan buatan di PLN UID Jawa Barat dalam meningkatkan efisiensi operasional, mengurangi kerugian finansial, dan memperkuat tata kelola risiko operasional, serta mendukung pilar Digital Moonshot dalam program Transformasi 2.0 PLN menuju visi Indonesia Emas 2045.
=====================================================================================================================================
Electricity theft is a serious issue that significantly impacts economic losses, operational efficiency, and the financial stability of utility companies. PT PLN (Persero) Distribution Unit of West Java (UID Jawa Barat), as the largest distribution unit in Indonesia, faces substantial challenges in managing this issue. Traditional detection approaches have been limited to subjective assumptions based on energy consumption analyses without a strategic prioritization for direct customer site investigations.
This study aims to optimize electricity theft detection at PLN UID Jawa Barat using Automated Machine Learning (AutoML) with multivariable and cost-sensitive learning approaches. Selected variables encompass customer behavioral features, network technical data, and geographical and socioeconomic attributes derived from over 6.4 million real customer inspection records collected from 2019 to 2023. Fourteen classification algorithms, including Gradient Boosting, Decision Tree, KNN, Logistic Regression, and Random Forest, were evaluated using stratified cross-validation. The Extreme Gradient Boosting Classifier (XGBoost) model demonstrated the highest performance, achieving an F1-score of 0.92 and an AUC of 0.87 on the holdout dataset, detecting 93% of cases with a precision rate of 93%. Implementing cost-sensitive learning, which assigns different penalties to classification errors, significantly improved the number of valid detection cases (true positives) from 22,883 to 75,884 customers after optimization, without increasing error rates. The optimized model has been integrated into PLN UID Jawa Barat’s operational system and internal EPM application. The automated pipeline developed not only accelerates analysis but also supports data-driven decision-making regarding field investigation priorities. A cost-benefit analysis indicated potential savings from reduced electricity theft amounting to IDR 334 billion, with a return on investment (ROI) of 13.57%. A comprehensive risk analysis was conducted to ensure model implementation security through periodic validation, data security integration, and user training. Overall, this study underscores the effectiveness of AI-based digital transformation in PLN UID Jawa Barat in enhancing operational efficiency, reducing financial losses, strengthening operational risk governance, and supporting the Digital Moonshot pillar of PLN’s Transformation 2.0 program towards achieving Indonesia’s Golden Vision 2045.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Automated Machine Learning, Cost-Sensitive Learning, Pencurian Tenaga Listrik, PLN, Stratified K-Fold Cross Validation, Electricity Theft
Subjects:	T Technology > T Technology (General) > T57.6 Operations research--Mathematics. Goal programming
Divisions:	Interdisciplinary School of Management and Technology (SIMT) > 61101-Master of Technology Management (MMT)
Depositing User:	Dino Arla
Date Deposited:	30 Jul 2025 04:07
Last Modified:	30 Jul 2025 04:07
URI:	http://repository.its.ac.id/id/eprint/123175

Actions (login required)

View Item