Pengembangan MLSecOps Framework Machine learning Berkelanjutan: Integrasi MLOps dan DevSecOps

Saputra, Adi (2026) Pengembangan MLSecOps Framework Machine learning Berkelanjutan: Integrasi MLOps dan DevSecOps. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

Text 7032222002-Doctoral.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Machine learning pada awalnya berfokus pada peningkatan akurasi prediksi dan efisiensi operasional karena keberhasilan model umumnya diukur berdasarkan kedua aspek tersebut. Namun, meningkatnya ancaman terhadap machine learning menuntut pengembangan yang tidak hanya berorientasi pada efisiensi, tetapi juga mengintegrasikan aspek keamanan agar solusi dapat diimplementasikan secara optimal dalam lingkungan industri. MLOps meningkatkan efisiensi pengembangan, namun belum sepenuhnya mengintegrasikan keamanan, sehingga masih rentan terhadap serangan seperti data poisoning, model drift, dan adversarial attacks. Sebaliknya, DevSecOps menyediakan pendekatan keamanan yang komprehensif, tetapi belum dirancang untuk karakteristik machine learning yang berbasis data dan model dinamis. Penelitian ini mengusulkan framework Machine learning Security Operations (MLSecOps) sebagai integrasi MLOps dan DevSecOps berbasis Security by Design, khususnya pada sektor telekomunikasi. Evaluasi menggunakan model Cross-Efficiency Data Envelopment Analysis (CE-DEA) menunjukkan bahwa MLSecOps mendominasi peringkat efisiensi optimal dengan skor 0.9068 hingga 0.9835, secara signifikan lebih tangguh dibandingkan metode tradisional yang anjlok pada rentang inefisiensi 0.2809 hingga 0.3300. Faktor penerimaan dianalisis menggunakan Structural Equation Modeling (SEM) dengan integrasi Technology Acceptance Model (TAM), the Theory of Planned Behavior (TPB), and Flow Theory. Hasil menunjukkan Subjective Norms berpengaruh signifikan terhadap niat adopsi, sementara enjoyment tidak signifikan terhadap Perceived Ease of Use, menunjukkan bahwa MLSecOps dipersepsikan sebagai alat kerja kritikal yang berfokus pada efisiensi proses development dan ketahanan model (robustness). Penelitian ini menghasilkan framework MLSecOps yang meningkatkan keamanan dan robustness model machine learning sekaligus mendukung pengembangan berkelanjutan melalui prinsip reusability dan reproducibility.
======================================================================================================================================
Machine learning has traditionally focused on improving predictive accuracy and operational efficiency, as model performance has generally been evaluated based on these two criteria. However, the increasing security threats targeting machine learning systems require development approaches that not only emphasize efficiency but also integrate security to ensure reliable deployment in industrial environments. While Machine Learning Operations (MLOps) improves development efficiency, it does not fully integrate security throughout the machine learning lifecycle, leaving development pipelines vulnerable to attacks such as data poisoning, model drift, and adversarial attacks. Conversely, DevSecOps provides comprehensive security practices for software development but was not specifically designed to address the data-driven and dynamic characteristics of machine learning. This study proposes a Machine Learning Security Operations (MLSecOps) framework based on Security by Design by integrating MLOps and DevSecOps, with a particular focus on the telecommunications sector. The proposed framework was evaluated using the Cross-Efficiency Data Envelopment Analysis (CE-DEA) model, which demonstrated that MLSecOps achieved superior operational efficiency, with efficiency scores ranging from 0.9068 to 0.9835, significantly outperforming conventional approaches that recorded efficiency scores between 0.2809 and 0.3300. User acceptance factors were analyzed using Structural Equation Modeling (SEM) by integrating the Technology Acceptance Model (TAM), the Theory of Planned Behavior (TPB), and Flow Theory. The results indicate that Subjective Norms have a significant positive effect on the intention to adopt MLSecOps, whereas Enjoyment has no significant effect on Perceived Ease of Use. These findings suggest that MLSecOps is perceived as a critical framework that enhances development efficiency while improving the robustness of machine learning models against security threats. This study contributes an MLSecOps framework that strengthens the security and robustness of machine learning systems while promoting sustainable machine learning development through the principles of reusability and reproducibility.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Machine learning, MLSecOps, DevSecOps, Data Envelopment Analysis, SEM
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.585 Cloud computing. Mobile computing. Q Science > QA Mathematics > QA76.754 Software architecture. Computer software Q Science > QA Mathematics > QA76.9.A25 Computer security. Digital forensic. Data encryption (Computer science) T Technology > T Technology (General) > T58.6 Management information systems T Technology > T Technology (General) > T59.7 Human-machine systems.
Divisions:	Interdisciplinary School of Management and Technology (SIMT) > 66105-Doctor of Technology Management (DMT)
Depositing User:	Adi Saputra
Date Deposited:	28 Jul 2026 02:38
Last Modified:	28 Jul 2026 02:38
URI:	http://repository.its.ac.id/id/eprint/138515

Actions (login required)

View Item