Optimasi Model LightGBM Dengan Cost Sensitive Learning Untuk Peningkatan Performa Deteksi Pencurian Listrik Di PLN Unit Induk Distribusi Jawa Barat

Septiyanto, Rafdhi Fatoni (2026) Optimasi Model LightGBM Dengan Cost Sensitive Learning Untuk Peningkatan Performa Deteksi Pencurian Listrik Di PLN Unit Induk Distribusi Jawa Barat. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221079-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Pencurian tenaga listrik menjadi permasalahan krusial yang menyebabkan kerugian ekonomi signifikan dan menurunkan efisiensi operasional di PT PLN (Persero) UID Jawa Barat. Pendekatan konvensional yang reaktif terbukti belum efektif dalam menanggulangi masalah kompleks ini. Penelitian ini bertujuan mengembangkan model deteksi pencurian listrik berbasis machine learning yang presisi dan adaptif dengan memanfaatkan data historis pelanggan. Algoritma Light Gradient Boosting Machine (LightGBM) diterapkan sebagai model klasifikasi utama dan Extreme Gradient Boosting (XGBoost) sebagai model dasar (baseline model) untuk mengidentifikasi pelanggan berpotensi melakukan pelanggaran. Optimalisasi performa model dilakukan melalui pendekatan cost-sensitive learning untuk menyeimbangkan konsekuensi biaya dari setiap jenis kesalahan klasifikasi. Proses penyesuaian hyperparameter dieksekusi menggunakan metode optimasi parameter guna menemukan kombinasi parameter paling efektif. Validasi model selanjutnya memanfaatkan teknik Stratified K-Fold Cross Validation untuk memastikan konsistensi hasil dan memitigasi bias akibat distribusi data yang tidak seimbang. Kinerja model dievaluasi secara komprehensif menggunakan berbagai metrik, termasuk confusion matrix, akurasi, precision, recall, F1-score, serta AUC untuk mengukur keandalannya dilanjutkan dengan cost benefit analysis. Penelitian ini diharapkan menghasilkan sistem deteksi berbasis data yang mampu memberikan nilai prediktif tinggi sebagai dasar pengambilan keputusan investigasi lapangan. Implementasi sistem ini juga diharapkan dapat menekan kerugian, memperkuat keandalan distribusi, serta mengakselerasi transformasi digital di industri ketenagalistrikan.
==================================================================================================================================
Electricity theft has become a crucial problem that causes significant economic losses and reduces operational efficiency at PT PLN (Persero) UID West Java. Conventional reactive approaches have proven ineffective in addressing this complex issue. This study aims to develop a precise and adaptive electricity theft detection model by utilizing customer historical data. The Light Gradient Boosting Machine (LightGBM) algorithm is applied as the main classification model and Extreme Gradient Boosting (XGBoost) as the baseline model to identify customers with potential violations. Model performance optimization is carried out through a Cost-Sensitive Learning approach to balance the cost consequences of each type of classification error. Hyperparameter tuning is executed using parameter optimization methods to find the most effective parameter combinations. Model validation further employs the Stratified K-Fold Cross Validation technique to ensure result consistency and mitigate bias due to imbalanced data distribution. The model’s performance is comprehensively evaluated using various metrics, including confusion matrix, accuracy, precision, recall, F1-score, and AUC to measure its reliability and cost benefit analysis. This research is expected to produce a data-driven detection system capable of providing high predictive value as a basis for field investigation decision-making. The implementation of this system is also expected to reduce losses, strengthen distribution reliability, and accelerate digital transformation in the electricity industry.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Big Data, LightGBM, Machine Learning, Pencurian Tenaga Listrik, Perusahaan Listrik Negara, Big Data, Electricity Theft, LightGBM, Machine Learning, State Electricity Company
Subjects:	Q Science Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Rafdhi Fatoni Septiyanto
Date Deposited:	29 Jan 2026 09:52
Last Modified:	29 Jan 2026 09:52
URI:	http://repository.its.ac.id/id/eprint/131157

Actions (login required)

View Item