Perancangan dan Implementasi Pipeline Machine Learning Operations (MLOps) Untuk Operasionalisasi Model Prediktif Risiko Kredit

Imanuel, Daulat Satrio Utomo Siahaan (2026) Perancangan dan Implementasi Pipeline Machine Learning Operations (MLOps) Untuk Operasionalisasi Model Prediktif Risiko Kredit. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5024221066-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Penelitian ini merancang dan mengimplementasikan pipeline Machine Learning Operations (MLOps) untuk sistem Credit Risk Scoring. Dalam konteks penelitian ini, operasionalisasi didefinisikan sebagai proses transisi model prediktif dari sebuah konsep matematika yang abstrak menjadi sistem perangkat lunak yang nyata, terukur, dan dapat disimulasikan penerapannya pada alur operasional lembaga keuangan. Model prediktif dilatih menggunakan algoritma XGBoost pada dataset sintetis berjumlah 10.000 sampel yang merepresentasikan profil nasabah ritel, yang diproses melalui transformasi Weight of Evidence (WoE) dan Information Value (IV). Untuk mengoperasionalisasikan model tersebut, pipeline MLOps dibangun mengintegrasikan pelacakan eksperimen, deployment antarmuka API, hingga otomasi siklus pelatihan ulang (Continuous Training) yang dipicu secara terjadwal maupun berbasis ketersediaan data baru. Hasil pengujian menunjukkan bahwa pendekatan ini tidak hanya menghasilkan model yang akurat dan dapat diinterpretasikan melalui SHAP, tetapi juga membuktikan bagaimana suatu algoritma cerdas dapat dioperasionalisasikan menjadi layanan yang adaptif, memiliki rekam jejak (audit trail) yang jelas, dan senantiasa terjaga kebaruannya untuk mendukung pengambilan keputusan.
====================================================================================================================================
This study designs and implements a Machine Learning Operations (MLOps) pipeline for a Credit Risk Scoring system. In the context of this research, operationalization is defined as the process of transitioning a predictive model from an abstract mathematical concept into a tangible, measurable software system whose application can be simulated within a financial institution’s operational workflow. The predictive model is trained using the XGBoost algorithm on a synthetic dataset of 10,000 samples representing retail borrower profiles, which are processed through Weight of Evidence (WoE) transformation and Information Value (IV) feature selection. To operationalize the model, the MLOps pipeline is built by integrating experiment tracking, API deployment, and an automated Continuous Training cycle triggered by scheduled intervals or the availability of new data. The evaluation results indicate that this approach not only produces an accurate and interpretable model through SHAP, but also demonstrates how an intelligent algorithm can be operationalized into an adaptive service with a clear audit trail, while consistently maintaining model freshness to support decision-making.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	credit risk scoring, MLOps, XGBoost, WoE, SHAP Credit risk scoring, MLOps, XGBoost, WoE, SHAP
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Computer Engineering > 90243-(S1) Undergraduate Thesis
Depositing User:	Imanuel Daulat Satrio Utomo Siahaan
Date Deposited:	21 Jul 2026 09:38
Last Modified:	21 Jul 2026 09:38
URI:	http://repository.its.ac.id/id/eprint/136031

Actions (login required)

View Item