Ekstraksi Wawasan Melalui Visualisasi Data dari Data Telekomunikasi PT. Telekomunikasi Selular

Putra, Davin Fisabilillah Reynard (2025) Ekstraksi Wawasan Melalui Visualisasi Data dari Data Telekomunikasi PT. Telekomunikasi Selular. Project Report. [s.n.], [s.l.]. (Unpublished)

Text 5025221137-Project_Report.pdf - Accepted Version Restricted to Repository staff only Download (1MB) | Request a copy

Abstract

Kegiatan kerja praktik di PT. Telekomunikasi Selular (Telkomsel) berfokus pada pemanfaatan data telekomunikasi untuk menghasilkan wawasan bisnis. Penulis ditempatkan di tim Insight as a Service (IaaS) pada posisi Product & GTM Strategic Alignment. Tujuan utama kerja praktik ini adalah untuk membantu tim delivery dalam proses pengolahan data telekomunikasi mentah menjadi actionable insights bagi klien B2B. Metodologi yang diterapkan mencakup analisis data, visualisasi data, dan ekstraksi wawasan. Proses ini memanfaatkan tools teknis seperti Python (dengan library Matplotlib dan Seaborn) dan Microsoft Excel (Pivot Table, Pivot Chart). Selama kegiatan kerja praktik, penulis berhasil menyelesaikan beberapa proyek analisis, termasuk analisis kompetitor untuk event organizer dan perusahaan taksi, penyusunan insight report untuk klien dari berbagai industri (seperti ride-hailing dan furniture), serta laporan people movement untuk instansi pemerintah. Hasil dari kerja praktik ini adalah pemahaman praktis dan mendalam tentang bagaimana data dapat ditransformasi menjadi rekomendasi strategis yang dalam bisnis.
Penulis juga mengembangkan model machine learning untuk klasifikasi skor kredit (credit scoring). Metodologi yang diterapkan mencakup analisis data eksploratif, preprocessing data, dan pembangunan model klasifikasi menggunakan algoritma Logistic Regression, Decision Tree, Random Forest, XGBoost, dan CatBoost. Pengujian dilakukan dalam tiga skenario: model baseline, optimasi hyperparameter menggunakan Optuna, dan penanganan imbalanced data dengan metode SMOTE. Hasil penelitian menunjukkan bahwa strategi hyperparameter tuning memberikan dampak paling signifikan, di mana model XGBoost berhasil mencapai performa terbaik dengan akurasi 77,45%, mengungguli model baseline dan penggunaan SMOTE yang justru menurunkan kinerja model berbasis boosting. Sementara itu, Random Forest terbukti sebagai model dengan stabilitas tertinggi di seluruh skenario pengujian.
=====================================================================================================================================
The internship activity at PT. Telekomunikasi Selular (Telkomsel) focused on the utilization of telecommunication data to generate business insights. The author was placed in the Insight as a Service (IaaS) team in the Product & GTM Strategic Alignment position. The main objective of this internship was to assist the delivery team in processing raw telecommunication data into actionable insights for B2B clients. The methodology applied included data analysis, data visualization, and insight extraction. This process utilized technical tools such as Python (with Matplotlib and Seaborn libraries) and Microsoft Excel (Pivot Table, Pivot Chart). During the internship activity, the author successfully completed several analysis projects, including competitor analysis for event organizers and taxi companies, preparation of insight reports for clients from various industries (such as ride-hailing and furniture), as well as people movement reports for government agencies. The result of this internship is a practical and in-depth understanding of how data can be transformed into strategic recommendations in business. The author also developed a machine learning model for credit score classification (credit scoring). The methodology applied included exploratory data analysis, data preprocessing, and classification model building using Logistic Regression, Decision Tree, Random Forest, XGBoost, and CatBoost algorithms. Testing was conducted in three scenarios: baseline model, hyperparameter optimization using Optuna, and imbalanced data handling using the SMOTE method. The research results showed that the hyperparameter tuning strategy provided the most significant impact, where the XGBoost model successfully achieved the best performance with an accuracy of 77.45%, outperforming the baseline model and the use of SMOTE which actually decreased the performance of boosting-based models. Meanwhile, Random Forest proved to be the model with the highest stability across all testing scenarios

Item Type:	Monograph (Project Report)
Uncontrolled Keywords:	Ekstraksi Wawasan, Visualisasi Data, Insight Report, Machine Learning, Hyperparameter Tuning, SMOTE, Insight Extraction, Data Visualization, Insight Report, Machine Learning, Hyperparameter Tuning, SMOTE.
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.9.D343 Data mining. Querying (Computer science) Q Science > QA Mathematics > QA76.9.I52 Information visualization
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Davin Fisabilillah Reynard Putra
Date Deposited:	19 Dec 2025 01:49
Last Modified:	19 Dec 2025 01:49
URI:	http://repository.its.ac.id/id/eprint/129054

Actions (login required)

View Item