Deteksi Penipuan Transaksi E-Commerce Menggunakan Integrasi Autoencoder Dan LightGBM Dengan Bayesian Optimization

Ridho, Achmad Fahmi Ainur (2026) Deteksi Penipuan Transaksi E-Commerce Menggunakan Integrasi Autoencoder Dan LightGBM Dengan Bayesian Optimization. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5026221167-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Pertumbuhan transaksi e-commerce meningkatkan risiko penipuan yang memiliki pola kompleks, tidak linear, serta rasio ketidakseimbangan kelas yang tinggi, sehingga diperlukan pendekatan deteksi yang efektif dan adaptif. Penelitian ini mengusulkan dan mengevaluasi pemanfaatan Autoencoder sebagai mekanisme latent-space oversampling yang diintegrasikan dengan Light Gradient Boosting Machine (LightGBM) untuk mendeteksi penipuan transaksi pada dataset IEEE-CIS Fraud Detection. Autoencoder dilatih pada sampel kelas penipuan untuk membangkitkan sampel sintetis melalui interpolasi pada ruang laten, sedangkan LightGBM digunakan sebagai model klasifikasi karena kinerjanya yang baik pada data tabular berskala besar. Sebagai pembanding terkendali digunakan SMOTE dengan rasio target yang sama, sementara dimensi ruang laten dan rasio target ditetapkan melalui analisis sensitivitas pada data validasi. Optimasi hyperparameter dilakukan menggunakan Bayesian Optimization berbasis Tree-structured Parzen Estimator (TPE) melalui Optuna secara terpisah pada setiap skenario, dengan Precision-Recall AUC (PR-AUC) sebagai metrik utama serta Recall, Precision, F1-Score, dan ROC-AUC sebagai metrik pendukung. Hasil pengujian menunjukkan bahwa setelah optimasi, model usulan memperoleh PR-AUC sebesar 0,8547, lebih tinggi daripada model dasar LightGBM (0,8497) dan model dengan SMOTE (0,8500). Dengan demikian, Autoencoder terbukti efektif sebagai latent-space oversampler yang meningkatkan kinerja deteksi penipuan pada representasi fitur padat.
======================================================================================================================================
The rapid growth of e-commerce transactions has increased the risk of fraud characterized by complex, nonlinear patterns and a high class-imbalance ratio, requiring an effective and adaptive detection approach. This study proposes and evaluates the use of an Autoencoder as a latent-space oversampling mechanism integrated with the Light Gradient Boosting Machine (LightGBM) to detect transaction fraud on the IEEE-CIS Fraud Detection dataset. The Autoencoder is trained on the fraud class to generate synthetic samples through interpolation in the latent space, while LightGBM serves as the classifier owing to its strong performance on large-scale tabular data. SMOTE with the same target ratio is used as a controlled comparison, and the latent dimension and target ratio are determined through a sensitivity analysis on the validation set. Hyperparameter optimization is performed using Bayesian Optimization based on the Tree-structured Parzen Estimator (TPE) via Optuna, separately for each scenario, with Precision-Recall AUC (PR-AUC) as the primary metric and Recall, Precision, F1-Score, and ROC-AUC as supporting metrics. The results show that after optimization, the proposed model achieves a PR-AUC of 0.8547, higher than the LightGBM baseline (0.8497) and the SMOTE-based model (0.8500). These findings demonstrate that the Autoencoder is effective as a latent-space oversampler that improves fraud detection performance on dense feature representations.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deteksi Penipuan, E-commerce, Autoencoder, LightGBM, Bayesian Optimization, Fraud Detection, E-commerce, Autoencoder, LightGBM, Bayesian Optimization.
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. T Technology > T Technology (General) > T174 Technological forecasting T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	Achmad Fahmi Ainur Ridho
Date Deposited:	31 Jul 2026 07:59
Last Modified:	31 Jul 2026 07:59
URI:	http://repository.its.ac.id/id/eprint/140734

Actions (login required)

View Item