Analisis Kinerja Metode Deteksi Fraud Pada Data Transaksi Kredit

Setyara, Gery Febrian (2024) Analisis Kinerja Metode Deteksi Fraud Pada Data Transaksi Kredit. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025201151-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (5MB) | Request a copy

Abstract

Seiring dengan meningkatnya transaksi kredit digital, deteksi penipuan yang cepat dan akurat menjadi krusial untuk menjamin keamanan serta meminimalkan kerugian finansial bagi institusi keuangan. Mengingat kasus penipuan jauh lebih sedikit dibandingkan dengan transaksi yang tidak fraud, hal ini menimbulkan tantangan dalam penelitian. Penelitian ini mengevaluasi metode deteksi penipuan terbaik menggunakan algoritma pembelajaran mesin pada data transaksi kredit. Proses penelitian melibatkan beberapa tahap pre-processing seperti penghapusan missing values dan StandardScaler. Selain itu, penghilangan outlier menggunakan Isolation Forest juga diterapkan untuk meningkatkan kualitas data. Penelitian ini membandingkan beberapa performa pembelajaran mesin seperti Random Forest, Logistic Regression, dan XGBoost, dengan teknik oversampling seperti SMOTE, Random Over Sampler, ADASYN, serta teknik undersampling seperti Random Under Sampler, NearMiss, dan Tomek Link untuk mengatasi masalah ketidakseimbangan kelas, di mana 99.83% adalah Non-Fraud dan hanya 0.17% adalah Fraud. Hasil eksperimen menunjukkan bahwa kombinasi Random Forest dengan resampling ADASYN memberikan performa terbaik dengan F1-score 0.935, dengan peningkatan F1-score kurang lebih sebesar 27% dibandingkan dengan tanpa penanganan ketidakseimbangan data dan 5-10% dibandingkan menggunakan metode imbalance handling lainnya.

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

With the increase in digital credit transactions, rapid and accurate fraud detection is crucial to ensure security and minimize financial losses for financial institutions. Given that fraud cases are much fewer compared to non-fraudulent transactions, this poses a challenge in research. This study evaluates the best fraud detection methods using machine learning algorithms on credit transaction data. The research process involves several preprocessing stages such as missing value removal and StandardScaler. Additionally, outlier removal using Isolation Forest is also applied to improve data quality. This study compares the performance of several machine learning algorithms like Random Forest, Logistic Regression, and XGBoost, with oversampling techniques such as SMOTE, Random Over Sampler, ADASYN, and undersampling techniques like Random Under Sampler, NearMiss, and Tomek Link to address the class imbalance problem, where 99.83% is Fraud and only 0.17% is Fraud. The experimental results show that the combination of Random Forest with ADASYN resampling provides the best performance with an F1-score of 0.935, with an increase in F1-score of approximately 27% compared to without handling data imbalance and 5-10% compared to using other imbalance handling techniques.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	ADASYN, credit transactions, Deteksi penipuan, Fraud detection, Ketidakseimbangan kelas, imbalanced class, Isolation Forest, Machine Learning, Pembelajaran Mesin, Random Forest, Transaksi kredit.
Subjects:	T Technology > T Technology (General) > T174 Technological forecasting
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Gery Febrian Setyara
Date Deposited:	01 Aug 2024 03:36
Last Modified:	01 Aug 2024 03:36
URI:	http://repository.its.ac.id/id/eprint/110542

Actions (login required)

View Item