Deteksi Fraud Pada Transaksi Kartu Kredit Menggunakkan Graphsage

Andriansyah, Muhammad Andriansyah (2025) Deteksi Fraud Pada Transaksi Kartu Kredit Menggunakkan Graphsage. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 50022111147-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (11MB) | Request a copy

Abstract

Pesatnya perkembangan teknologi digital telah mendorong peningkatan signifikan pada volume transaksi keuangan elektronik, khususnya transaksi menggunakan kartu
kredit. Namun, pertumbuhan ini turut disertai oleh peningkatan risiko aktivitas
penipuan (fraud). Pendekatan pembelajaran mesin konvensional yang hanya menganalisis
transaksi secara individual sering kali tidak mampu menangkap hubungan kompleks
antar-entitas,seperti pelanggan, merchant, dan transaksi itu sendiri. Untuk mengatasi
keterbatasan tersebut, Tugas Akhir ini menerapkan pendekatan Graph Representation
Learning (GRL) menggunakan model GraphSAGE (Graph Sample and Aggregate)
yang memungkinkan pembelajaran secara induktif, sehingga model dapat menghasilkan
representasi vektor untuk entitas baru tanpa perlu melakukan pelatihan ulang terhadap
keseluruhan model. Penelitian ini memodelkan transaksi kartu kredit dalam bentuk
graf heterogen yang terdiri dari dua jenis node, yaitu user dan merchant. Dataset
yang digunakan telah melalui proses balancing menggunakan teknik undersampling
agar distribusi label seimbang. Eksperimen dilakukan dengan menguji tiga jenis
aggregator pada GraphSAGE, yaitu mean aggregator, max pooling aggregator, dan
LSTM aggregator. Hasil evaluasi menunjukkan bahwa model dengan mean aggregator
memberikan performa terbaik dengan akurasi sebesar 75.1%, precision sebesar 75.1%,
recall sebesar 75.1%, dan F1-score sebesar 75.1%. Performa ini lebih unggul dibandingkan
LSTMaggregator (74,6%) dan max pooling aggregator (73,2%). Penelitian ini menegaskan
bahwa pendekatan Graph Representation Learning mampu menangkap pola anomali yang
tersembunyi dalam relasi antar-entitas, bukan sekadar karakteristik dari satu transaksi
saja. Oleh karena itu, penggunaan GraphSAGE direkomendasikan untuk sistem deteksi
fraud pada kartu kredit yang membutuhkan efisiensi dan kemampuan generalisasi pada
data baru.

===========================================================

The rapid growth of digital technology has led to a significant rise in electronic
f
inancial transactions, particularly credit-card payments. This expansion, however,
is accompanied by an increased risk of fraudulent activities. Conventional machine
learning approaches that analyze each transaction in isolation often fail to capture
the complex relationships among entities such as customers, merchants, and the
transactions themselves. To address this limitation, this thesis employs a Graph
Representation Learning (GRL) approach using the GraphSAGE (Graph Sample and
Aggregate) model, which supports inductive learning and thus can generate vector
representations for previously unseen entities without retraining the entire model.
Credit-card transactions are modeled as a heterogeneous graph comprising two node
types—users and merchants. The dataset is balanced through undersampling to ensure
an even label distribution. Experiments evaluate three GraphSAGE aggregators mean,
max pooling, and LSTM. Results show that the mean aggregator delivers the best
performance, achieving an accuracy of 75.1%, precision of 75.1%, recall of 75.1%, and
F1-score of 75.1%, outperforming the max-pooling (73.2%) and LSTM (74.6%) variants.
These findings confirm that Graph Representation Learning effectively captures hidden
anomaly patterns within inter-entity relationships rather than relying solely on individual
transaction features. Consequently, GraphSAGE particularly with the mean aggregator
is recommended for credit-card fraud-detection systems that require both efficiency and
strong generalization to new data.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Fraud, Kartu Kredit, GraphSAGE, Aggregator
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Andriansyah
Date Deposited:	01 Aug 2025 09:07
Last Modified:	01 Aug 2025 09:07
URI:	http://repository.its.ac.id/id/eprint/125906

Actions (login required)

View Item