Impression : Jurnal Teknologi dan Informasi
Vol. 5 No. 1 (2026): Maret 2026

Optimasi Random Forest untuk Peningkatan Sensitivitas Deteksi Fraud pada Data Imbalanced

Tammamah Lubis, Hartati (Unknown)
Harianto, Adi (Unknown)



Article Info

Publish Date
15 Feb 2026

Abstract

Deteksi fraud pada transaksi keuangan merupakan permasalahan krusial akibat ketidakseimbangan data (imbalanced dataset) dan tingginya risiko kerugian finansial. Penelitian ini bertujuan mengoptimasi model Random Forest dalam mendeteksi transaksi fraud menggunakan dataset credit.csv. Proses optimasi dilakukan melalui RandomizedSearchCV untuk memperoleh kombinasi hyperparameter terbaik serta penyesuaian threshold probabilitas untuk meningkatkan nilai recall. Hasil eksperimen menunjukkan model mencapai nilai ROC-AUC sebesar 0.97 +, dengan recall fraud meningkatkan dari 0.83 pada threshold default menjadi 0.90 pada threshold 0.2. Hasil ini menunjukkan bahwa optimasi hyperparameter dan threshold efektif dalam meningkatkan sensitivitas sistem terhadap transaksi fraud.   Financial transaction fraud detection is a critical problem due to class imbalance in the dataset and the high financial losses associated with undetected fraudulent activities. This study aims to optimize a Random Forest model for detecting fraudulent transactions using the creditcard.csv dataset. The optimization process was conducted using RandomizedSearchCV to identify the best hyperparameter combination, and a probability threshold adjustment to improve recall performance. Experimental results show that the optimized model achieved an ROC-AUC score above 0.97, with fraud recall increasing from 0.83 at the default threshold to 0.90 at a threshold of 0.2. These findings indicate that hyperparameter tuning and threshold optimization effectively enhance the system's sensitivity to detecting fraudulent transactions.

Copyrights © 2026






Journal Info

Abbrev

jti

Publisher

Subject

Automotive Engineering Chemical Engineering, Chemistry & Bioengineering Civil Engineering, Building, Construction & Architecture Computer Science & IT Electrical & Electronics Engineering

Description

Impression accepts articles in the fields of Electrical Engineering, Mechanical Engineering, Civil Engineering, Marine Technology Industrial Engineering, Marine Fisheries Technology, Agricultural Technology, Informatics Engineering, Information Systems, Computer, Expert systems, Decision Support ...