Integrasi Machine Learning Operations: Experiment Tracking, Monitoring Model, dan CI/CD di Perusahaan XYZ

Bayuputra, Rayhan Arvianta (2025) Integrasi Machine Learning Operations: Experiment Tracking, Monitoring Model, dan CI/CD di Perusahaan XYZ. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211217-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (16MB) | Request a copy

Abstract

Machine Learning Operations (MLOps) merupakan pendekatan sistematis yang menggabungkan prinsip DevOps dengan pembelajaran mesin untuk meningkatkan efisiensi dan keandalan pengelolaan model di lingkungan produksi. Di Perusahaan XYZ, yang bergerak di bidang penyediaan mesin konstruksi dan alat berat, penerapan MLOps menghadapi berbagai tantangan, mulai dari pelacakan eksperimen yang belum optimal, keterbatasan dalam monitoring performa model, hingga kurangnya otomatisasi pada pipeline CI/CD. Penelitian ini bertujuan untuk merancang dan mengimplementasikan pipeline MLOps berbasis layanan Google Cloud yang mencakup pelacakan eksperimen, orkestrasi pipeline, serta pemantauan performa model secara real-time. Sistem dibangun dengan MLflow untuk experiment tracking, Apache Airflow untuk penjadwalan inference, Evidently untuk monitoring performa dan data drift, serta Azure Pipelines dan Cloud Build untuk otomatisasi deployment. Evaluasi dilakukan pada dua skenario pembelajaran mesin, yaitu klasifikasi dan regresi, dengan menguji setiap komponen utama secara end-to-end. Hasil pengujian menunjukkan bahwa seluruh komponen sistem berhasil dijalankan dengan tingkat keberhasilan 100% pada kedua skenario. Setiap proses, mulai dari pelatihan model, registrasi ke MLflow, inference terjadwal, pemantauan performa dan drift, hingga promosi lintas lingkungan melalui CI/CD, telah tervalidasi dan menghasilkan keluaran data serta artefak yang terdokumentasi secara konsisten.
=================================================================================================================================
Machine Learning Operations (MLOps) is a systematic approach that combines DevOps principles with machine learning to enhance the efficiency and reliability of model management in production environments. At Company XYZ, a heavy equipment and construction machinery provider, the implementation of MLOps faces several challenges, including suboptimal experiment tracking, limited model performance monitoring, and the absence of CI/CD pipeline automation. This study aims to design and implement a cloud-based MLOps pipeline that supports model experimentation, pipeline orchestration, and real-time performance monitoring. The system is built using MLflow for experiment tracking, Apache Airflow for scheduled inference, Evidently for performance and drift monitoring, and Azure Pipelines with Cloud Build for deployment automation. The evaluation was conducted on two machine learning scenarios, classification and regression, by executing every pipeline component end-to-end. The evaluation results indicate a 100% success rate across both scenarios, with each process from model training, MLflow registration, and scheduled inference to performance and drift monitoring and environment promotion via CI/CD, are all executed consistently and producing expected outputs and documented artifacts.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	CI/CD, experiment tracking, Google Cloud , MLOps, model monitoring. CI/CD, experiment tracking, Google Cloud, MLOps, monitoring model
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.758 Software engineering Q Science > QA Mathematics > QA76.9.D37 Data warehousing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Rayhan Arvianta Bayuputra
Date Deposited:	31 Jul 2025 01:06
Last Modified:	31 Jul 2025 01:06
URI:	http://repository.its.ac.id/id/eprint/123311

Actions (login required)

View Item