Mitigating learning biases - A study on uncertainty quantification and instantial variability in machine learning models

Aug 5, 2026

Upload by Ezra

Machine learning (ML) enables advanced healthcare solutions, yet uncertainty and data/model limitations can undermine reliability. This thesis investigates how instantial variability affects accuracy and robustness of ML models in EEG microstate analysis, where brief brain activity patterns are used to study neural dynamics. Resting-state EEG recordings from 203 participants were analyzed using features such as Global Explained Variance, Mean Durations, and Corrected Time Coverage. Probabilistic augmentation and uncertainty-aware classification were evaluated with KNN, ACS, ACG, WSF, and compared to LSVM under baseline and perturbed conditions. Results highlight the value of variability-aware techniques for more robust clinical diagnostics and brain-computer interfaces, motivating future work on larger datasets and real-time methods.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages65

Reading Time2 h 44 min

AccessFree to download

1 Introduction

2 Background

2.1 Uncertainty Quantiﬁcation Methods

2.2 Instantial Variability

2.3 EEG Microstate Analysis

3 Methods

3.1 Dataset Description

3.1.1 Preprocessing

3.1.2 Microstates and Their Classiﬁcation

3.2 Simulating Variability in Microstates: Artiﬁcial Perturbations of Microstate Features

3.2.1 Simulating Perturbations

3.2.2 Quantifying Perturbations

3.3 Classiﬁcation Models with Implementation

3.4 Assessing the Impact of Variability on Classiﬁcation

3.4.1 Microstates and their artiﬁcial perturbation

3.4.2 Evaluation Metrics

4 Results and Discussions

4.1 Variability Analysis

4.2 The Impact of Variability on Classiﬁcation

4.2.1 Classiﬁcation Performance with no Perturbations

4.3 Performance Degradation under Perturbations

4.3.1 Comparative Analysis of Models

5 Conclusions and Future Works

Master Thesis in ICT FOR INTERNET AND MULTIMEDIA Mitigating learning biases: A study on uncertainty quantiﬁcation and instantial variability in machine learning models Master Candidate Supervisor Alda Kola Prof. Roberto Corvaja Student ID 2071530 University of Padova Co-supervisor Prof. Giulia Cisotto University of Trieste Academic Year: 2023/2024 Graduation Date: December, 4th 2024 To my family. To my parents, Valentin and Dila, to all their sacriﬁces. To my sister Florinda,, my absolute favorite person, who has always been my light. Abstract Machine learning (ML) has become a signiﬁcant driver of advancement in healthcare, enabling advanced solutions for diagnostics, personalized treatments, and decision-making support. In the study of brain dynamics, EEG microstate analysis stands out as a valuable technique for exploring brain activity, oﬀering insights into cognitive functions and neural processes. However, inherent data variability and model limitations introduce uncertainty, posing signiﬁcant challenges to the reliability of ML applications in this domain. This thesis explores the impact of instantial variability on the accuracy and robustness of machine learning (ML) models in EEG microstate analysis. EEG microstates, brief patterns of brain activity, are crucial for understanding neural dynamics. Using a dataset of resting-state EEG recordings from203participants, key microstate features such as Global Explained Variance, Mean Durations, and Corrected Time Coverage were analyzed. To simulate variability, probabilistic augmentation techniques were applied on the dataset and uncertainty-aware methods were used to classify microstates. Four classiﬁers: K-Nearest Neighbors (KNN), Augmented Support Vector Classiﬁer (ACS), Augmented Gradient Boosting Classiﬁer (ACG), and Weighted Sampling Forest (WSF) were evaluated under baseline and perturbed conditions and compared with the performance of a traditional ML model, Linear Support Vector Machine (LSVM). The ACS model consistently showed the highest performance, demonstrating the eﬀectiveness of augmentation and uncertainty quantiﬁcation in enhancing robustness. To further evaluate the robustness of these classiﬁers, perturbations were introduced to simulate real-world variability. The ﬁndings emphasize the importance of variability-aware techniques in improving ML models for EEG analysis, paving the way for more reliable applications in clinical diagnostics and brain-computer interfaces. Future work should focus on expanding datasets, exploring deep learning approaches, and adapting methods to real-time applications. Contents List of Figures xi List of Tables xiii List of Acronyms xix 1 Introduction 1 2 Background 5 2.1 Uncertainty Quantiﬁcation Methods ................. 7 2.2 Instantial Variability .......................... 8 2.3 EEG Microstate Analysis ........................ 10 3 Methods 13 3.1 Dataset Description ........................... 14 3.1.1 Preprocessing .......................... 14 3.1.2 Microstates and Their Classiﬁcation . . . . . . . . . . . . . 15 3.2 Simulating Variability in Microstates: Artiﬁcial Perturbations of Microstate Features ........................... 16 3.2.1 Simulating Perturbations ................... 16 3.2.2 Quantifying Perturbations ................... 17 3.3 Classiﬁcation Models with Implementation ............. 19 3.4 Assessing the Impact of Variability on Classiﬁcation ........ 21 3.4.1 Microstates and their artiﬁcial perturbation ......... 22 3.4.2 Evaluation Metrics . . . . . . . . . . . . . . . . . . . . . . . 23 4 Results and Discussions 25 4.1 Variability Analysis ........................... 26 4.2 The Impact of Variability on Classiﬁcation .............. 30 CONTENTS 4.2. 1 Classiﬁcation Performance with no Perturbations . . . . . 31 4.3 Performance Degradation under Perturbations ........... 35 4.3.1 Comparative Analysis of Models ..........

FAQs

What is the focus of the thesis on learning biases?

The thesis studies how learning biases manifest through uncertainty and instantial variability, and how these factors impact ML performance in EEG microstate analysis.

Which EEG microstate features and data are used in the study?

The work analyzes resting-state EEG recordings from 203 participants and uses microstate features including Global Explained Variance, Mean Durations, and Corrected Time Coverage.

How is variability simulated and how are classifiers evaluated?

Probabilistic augmentation and uncertainty-aware methods classify microstates, with perturbations introduced to simulate real-world variability. KNN, ACS, ACG, and WSF are compared against a baseline LSVM model under both baseline and perturbed conditions.

Aug 5, 2026

Upload by Ezra