Implementasi Explainable Ai Pada Model Light Gradient Boosting Machine Untuk Memprediksi Churn Pada Industri Telekomunikasi (Studi Kasus PT.XYZ)

Fernando, Edbert (2026) Implementasi Explainable Ai Pada Model Light Gradient Boosting Machine Untuk Memprediksi Churn Pada Industri Telekomunikasi (Studi Kasus PT.XYZ). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2043221106-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Fenomena customer churn menjadi tantangan utama bagi industri telekomunikasi karena biaya akuisisi pelanggan baru jauh lebih besar dibandingkan mempertahankan pelanggan lama. Penurunan basis pelanggan berdampak langsung pada kinerja finansial, sehingga diperlukan metode prediksi yang akurat sekaligus transparan bagi pemangku kepentingan. Penelitian ini mengusulkan implementasi Explainable Artificial Intelligence (XAI) pada algoritma Light Gradient Boosting Machine (LGBM) untuk memprediksi churn di PT. XYZ. Metodologi penelitian meliputi tahapan preprocessing, encoding, dan penanganan imbalanced data menggunakan metode SMOTENC. Kinerja model dievaluasi berdasarkan metrik akurasi, precision, recall, F1-score, dan ROC-AUC. Untuk meningkatkan transparansi, diterapkan metode SHAP (SHapley Additive exPlanations) untuk interpretasi global dan LIME (Local Interpretable Model-agnostic Explanations) untuk interpretasi lokal. Hasil penelitian menunjukkan bahwa model LightGBM mampu memberikan performa tinggi dengan nilai ROC-AUC sebesar 0,8105 dan recall 61,76%. Analisis SHAP mengungkap bahwa faktor paling dominan yang memengaruhi churn adalah Tenure Months dan Device Class, sementara LIME berhasil menjelaskan alasan prediksi pada level individu pelanggan secara logis. Integrasi ini mendukung pengambilan keputusan strategis yang lebih terukur dalam manajemen retensi pelanggan.
===============================================================================================================================
Customer churn represents a critical challenge in the telecommunications industry, as the cost of acquiring new customers significantly exceeds that of retaining existing ones. A declining customer base directly impacts financial performance, necessitating prediction methods that are both accurate and interpretable for stakeholders. This study proposes the implementation of Explainable Artificial Intelligence (XAI) using the Light Gradient Boosting Machine (LightGBM) algorithm to predict customer churn at PT. XYZ. The research methodology includes data preprocessing, categorical encoding, and addressing class imbalance via the SMOTENC technique. Model performance is evaluated using accuracy, precision, recall, F1-score, and ROC-AUC metrics. To enhance model transparency, XAI is implemented through the SHAP approach for global interpretability and LIME for local, instance-level explanations. The results demonstrate that the LightGBM model achieves high predictive performance with an ROC-AUC of 0.8105 and a recall of 61.76%. SHAP analysis identifies Tenure Months and Device Class as the most influential factors driving churn, while LIME successfully provides logical justifications for individual-level predictions. This integration offers a comprehensive framework that supports data-driven strategic decision-making in customer retention management.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Customer Churn, Explainable AI, LGBM, LIME, Machine Learning, SHAP, SMOTENC,
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence
Divisions:	Faculty of Vocational > 49501-Business Statistics
Depositing User:	Edbert Fernando
Date Deposited:	30 Mar 2026 05:28
Last Modified:	30 Mar 2026 05:28
URI:	http://repository.its.ac.id/id/eprint/132741

Actions (login required)

View Item