Pipeline Machine Learning Operations Berbasis Low-Code Untuk Deployment Dan Monitoring Model Deteksi Penipuan Kartu Kredit

Syahputra, Ahmad Wildan (2026) Pipeline Machine Learning Operations Berbasis Low-Code Untuk Deployment Dan Monitoring Model Deteksi Penipuan Kartu Kredit. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5024221064-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB)

Abstract

Peningkatan volume transaksi perbankan digital memperbesar risiko penipuan kartu kredit. Model deteksi berbasis *machine learning* yang memiliki kinerja baik pada tahap pelatihan dapat mengalami penurunan performa di lingkungan produksi akibat *concept drift*, sementara proses *deployment* secara manual cenderung lambat dan rentan terhadap kesalahan. Penelitian ini merancang dan mengimplementasikan *pipeline* Machine Learning Operations (MLOps) berbasis *low-code* menggunakan platform n8n untuk *deployment* dan pemantauan model deteksi penipuan kartu kredit berbasis XGBoost. Sistem mengintegrasikan *model registry* berbasis GitHub dengan mekanisme *hot-reload* untuk pembaruan model tanpa *downtime*, pencatatan prediksi untuk mendukung *observability*, serta deteksi *drift* menggunakan uji Kolmogorov–Smirnov untuk *data drift* dan uji Chi-Squared dengan *fallback* Fisher’s Exact untuk *prediction drift*. Kebaruan penelitian ini terletak pada skema peringatan dua tingkat (*2-Tier Alerting*) yang menerapkan ambang batas lebih ketat pada tiga fitur terpenting model untuk menekan alarm palsu akibat pengujian statistik berganda. Hasil pengujian menunjukkan bahwa sistem mampu melayani 960 dari 960 permintaan tanpa kegagalan selama proses pembaruan model, dengan waktu *hot-reload* sebesar 179,9 milidetik. Sistem juga berhasil mendeteksi *synthetic drift* tanpa menghasilkan alarm palsu pada data normal serta memangkas *deployment lead time* sebesar 96,6%, dari 21,36 detik menjadi 0,72 detik, dengan *median logging overhead* yang mendekati nol. Hasil penelitian ini menunjukkan bahwa pendekatan MLOps berbasis *low-code* layak diterapkan untuk meningkatkan keandalan dan efisiensi operasional sistem deteksi penipuan.
=================================================================================================================================
The growth of digital banking transactions has increased the risk of credit card fraud. A machine learning-based detection model that performs well during training may experience performance degradation in production due to concept drift, while manual deployment processes are often slow and prone to errors. This research designs and implements a low-code Machine Learning Operations (MLOps) pipeline using the n8n platform for the deployment and monitoring of an XGBoost-based credit card fraud detection model. The system integrates a GitHub-based model registry with a hot-reload mechanism for zero-downtime model updates, prediction logging to support observability, and statistical drift detection using the Kolmogorov–Smirnov test for data drift and the Chi-Squared test with Fisher’s Exact test as a fallback for prediction drift. The novelty of this research lies in a two-tier alerting scheme that applies stricter thresholds to the three most important model features to suppress false alarms resulting from multiple statistical tests. Experimental results show that the system successfully served all 960 requests without failure during model updates, achieving a hot-reload time of 179.9 milliseconds. The system also detected synthetic drift without generating false alarms on normal data and reduced deployment lead time by 96.6%, from 21.36 seconds to 0.72 seconds, while maintaining a near-zero median logging overhead. These findings demonstrate that a low-code MLOps approach is a practical solution for improving the reliability and operational efficiency of credit card fraud detection systems.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Penipuan Kartu Kredit, Machine Learning Operations (MLOps), Low-Code, Concept Drift, Deployment Otomatis, Monitoring Model, Credit Card Fraud, Machine Learning Operations (MLOps), Low-Code, Concept Drift, Automated Deployment, Model Monitoring.
Subjects:	T Technology > T Technology (General) > T58.5 Information technology. IT--Auditing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Computer Engineering > 90243-(S1) Undergraduate Thesis
Depositing User:	Ahmad Wildan Syahputra
Date Deposited:	23 Jul 2026 03:29
Last Modified:	23 Jul 2026 03:29
URI:	http://repository.its.ac.id/id/eprint/136320

Available Versions of this Item

• Pipeline Machine Learning Operations Berbasis Low-Code Untuk Deployment Dan Monitoring Model Deteksi Penipuan Kartu Kre. (deposited 22 Jul 2026 06:41)
  • Pipeline Machine Learning Operations Berbasis Low-Code Untuk Deployment Dan Monitoring Model Deteksi Penipuan Kartu Kredit. (deposited 23 Jul 2026 03:29) [Currently Displayed]

Actions (login required)

View Item