Optimasi Hiperparameter pada Tree-based Models Menggunakan Algoritma Genetika untuk Prediksi Pendapatan Layanan SMS Internasional

Farokhi, Jiryan (2026) Optimasi Hiperparameter pada Tree-based Models Menggunakan Algoritma Genetika untuk Prediksi Pendapatan Layanan SMS Internasional. Other thesis, Insitut Teknologi Sepuluh Nopember.

Text 5002221102-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Industri telekomunikasi menghadapi tantangan dalam memprediksi pendapatan layanan SMS internasional karena trafik, rute, pelanggan, biaya, dan pola musiman membentuk hubungan yang dinamis serta nonlinier. Pada saat yang sama, pendapatan
SMS internasional juga tertekan oleh fenomena revenue leakage, seperti gray routes, Flash Calls, dan alternatif otentikasi yang lebih murah. Meskipun berbagai penelitian telah menunjukkan bahwa Algoritma Genetika efektif digunakan untuk optimasi hiperparameter pada model machine learning, penelitian tersebut belum ada yang secara khusus mengkaji prediksi pendapatan layanan SMS internasional. Penelitian
Tugas Akhir ini menerapkan Algoritma Genetika untuk mengoptimalkan hiperparameter model berbasis pohon pada prediksi TotalRevenue layanan SMS internasional perusahaan
X. Dataset primer yang digunakan merupakan data transaksi SMS A2P tahun 2023 sampai 2024 sebanyak 94.335 baris dengan 19 kolom. Data dibersihkan, disamarkan untuk menjaga kerahasiaan, diagregasi secara bulanan berdasarkan Series_ID, kemudian direkayasa menjadi fitur historis, operasional, anomali, dan temporal. Pemodelan dilakukan dengan pendekatan dua tahap, yaitu klasifikasi transaksi pendapatan tidak nol dan pendapatan nol menggunakan XGBoost, kemudian regresi pendapatan untuk transaksi yang diprediksi memiliki pendapatan tidak nol menggunakan Random Forest dan XGBoost. Optimasi hiperparameter dijalankan menggunakan Algoritma Genetika dengan populasi 40 individu, 75 generasi, crossover rate 1,0, mutation rate 0,2, dan keep_elitism 4 individu. Hasil penelitian menunjukkan bahwa optimasi hiperparameter meningkatkan performa kedua model. WMAPE Random Forest turun dari 25,37% menjadi 20,99%, sedangkan WMAPE XGBoost turun dari 27,12% menjadi 16,29%. Model XGBoost dengan Algoritma Genetika memberikan hasil terbaik dengan WMAPE 16,29%, MAE 23.398,93, dan R2 0,9765.
==========================================================================================================================================
The telecommunications industry faces challenges in predicting international SMS service revenue because traffic, routes, customers, costs, and seasonal patterns form dynamic and nonlinear relationships. At the same time, international SMS revenue is also pressured by revenue leakage phenomena, such as gray routes, Flash Calls, and cheaper authenticationalternatives. Although various studies have shown that GeneticAlgorithms are effective for hyperparameter optimization in machine learning models, no study has specifically examined revenue prediction for international SMS services. This Final Project applies a Genetic Algorithm to optimize the hyperparameters of tree-based models for predicting the TotalRevenue of international SMS services at Company X.The primary dataset used consists of A2P SMS transaction data from 2023 to 2024, containing 94,335 rows and 19 columns. The data were cleaned, anonymized to preserve confidentiality, aggregated monthly based on Series_ID, and then engineered into historical, operational, anomaly-based, and temporal features. The modeling process was conducted using a two-stage approach: classification of nonzero-revenue and zero-revenue transactions using XGBoost, followed by revenue regression for transactions predicted to have nonzero revenue using Random Forest and XGBoost. Hyperparameter optimization was performed using a Genetic Algorithm with a population of 40 individuals, 75 generations, a crossover rate of 1.0, a mutation rate of 0.2, and keep_elitism of 4 individuals. The results show that hyperparameter optimization improved the performance of both models. The WMAPE of Random Forest decreased from 25.37% to 20.99%, while the WMAPE of
XGBoost decreased from 27.12% to 16.29%. The XGBoost model optimized with the Genetic Algorithm achieved the best performance, with a WMAPE of 16.29%, an MAE of 23,398.93, and an R2 of 0.9765.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Pendapatan Layanan SMS Internasional, Tree-based Models, Optimasi Hiperparameter, Algoritma Genetika, Revenue of SMS International Services, Tree-based Models, Hyperparameter Optimization, Genetic Algorithm
Subjects:	Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) Q Science > QA Mathematics > QA278.2 Regression Analysis. Logistic regression Q Science > QA Mathematics > QA402.5 Genetic algorithms. Interior-point methods. Q Science > QA Mathematics > QA76.9.D343 Data mining. Querying (Computer science) Q Science > QA Mathematics > QA9.58 Algorithms
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Jiryan Farokhi
Date Deposited:	31 Jul 2026 08:21
Last Modified:	31 Jul 2026 08:21
URI:	http://repository.its.ac.id/id/eprint/141032

Actions (login required)

View Item