Operationalizing Machine Learning for Healthcare: A Hybrid Framework for Data Drift Detection an‎d Adaptive Model Updating in Medical Imaging

4/21/2026 12:00:00 AM

1 Chapter 1: Introduction 13 1.1 Overview 14 1.2 Problem Statement 15 1.3 Contributions 16 1.4 Innovations 17 1.5 Thesis Structure 18 2 Chapter 2: Background an‎d Theoretical Concepts 20 2.1 General MLOps Architecture 21 2.1.1 Overview of MLOps Architecture 21 2.1.2 Advantages of MLOps 23 2.1.3 Applications of MLOps 24 2.1.4 MLOps Tools 25 2.1.5 Challenges in MLOps 26 2.2 Data Drift in Machine Learning 27 2.2.1 Definition an‎d Importance of Data Drift 27 2.2.2 Types of Data Drift (Covariate, Concept, Label Drift) 27 2.2.3 Causes of Data Drift (User Behaviour, Environment, System Changes) 28 2.2.4 Impact on Model Performance an‎d Reliability 29 2.3 Drift Detection an‎d Analysis 30 2.3.1 Statistical Methods for Drift Detection 30 2.3.2 Machine Learning-Based Detection Approaches 31 2.3.3 Hybrid Detection Strategies 33 2.3.4 eva‎luation Metrics for Drift Detection 34 2.4 Drift Mitigation an‎d Adaptive Responses 36 2.4.1 Model Retraining Strategies 36 2.4.2 Online Learning an‎d Incremental updat‎es 37 2.4.3 Feedback Loops for Continuous Adaptation 37 2.4.4 Integration into MLOps Pipelines 38 2.5 Summary 39 3 Chapter 3: Literature Review 40 3.1 Introduction 41 3.2 Overview on Previous work 41 3.2.1 Statistical an‎d Ensemble-Based Detection Methods 42 3.2.2 Autoencoder an‎d Deep Learning-Based Detection 43 3.2.3 Meta-Learning an‎d Task-Sensitive Approaches 44 3.2.4 Sliding Window an‎d Multi-Linear Approaches 46 3.2.5 Large-Scale System-Level Mitigation 46 3.2.6 Drift Detection in Business Process an‎d Event Logs 47 3.3 Comparative study 48 3.4 Summary an‎d Implications 50 4 Chapter 4: Methodology 53 4.1 Introduction 54 4.2 Overview of The Proposed Methodology 54 4.3 Stages of Methodology 56 4.4 Characteristics of the Framework 59 4.4.1 Statistical Monitoring Algorithms 59 4.4.2 Deep Learning–Based Prediction Auditing 60 4.4.3 Controlled Adaptation an‎d selec‎tive Review 60 4.5 Summary 61 5 Chapter 5: Experimental Results 62 5.1 Introduction 63 5.2 Experimental Setup 63 5.2.1 Dataset Description 64 5.2.2 Data Splitting an‎d Stream Batches 65 5.2.3 Exploratory Data Analysis (EDA) 66 5.2.4 Baseline Model Training 68 5.2.5 Feature Extraction an‎d Drift Reference 71 5.2.6 Drift Metrics Calculation 72 5.2.7 Hybrid Drift Score an‎d Drift al‎e‎rt 75 5.2.8 Adaptive Model updat‎e (Policy A & Policy B) 77 5.3 Results an‎d Discussion 79 5.3.1 Drift Detection Results 79 5.3.2 Batch-wise Analysis an‎d Drift Impact 79 5.3.3 The Effect of Adaptive Model updat‎es 80 5.3.4 Comparative eva‎luation with Existing Methods 80 5.3.5 Discussion 82 5.4 Summary 82 6 Chapter 6: Conclusion an‎d Future Work 84 6.1 Conclusion 85 6.2 Future Work 86 7 References 87

MLOps , رانش داده , رانش مفهوم

MLOps, , Data Drift , Concept Drift

Mustafa Naeem Obaid Obaid

Dr. Behrouz Minaei-Bidgoli