Towards a Comprehensive Model for Cognitive Performance Prediction - A Machine Learning-Based Approach - Thesis

Aug 3, 2026

Upload by Skyler

Advances in wearable technology enable data-driven insights into cognitive performance and mental well-being. This thesis presents a pilot study that integrates multimodal inputs, including nutrition, sleep, physical activity, and real-time physiological monitoring, to track and predict cognitive performance across memory, reasoning, attention, reading, and learning. A predictive framework is built with machine learning and deep learning models, enhanced interpretability via SHAP and occlusion sensitivity, and counterfactual analysis to derive actionable, personalized lifestyle recommendations. The work details collection pipelines, preprocessing strategies, and experimental outcomes, establishing a basis for real-time individualized cognitive enhancement.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages135

Reading Time5 h 40 min

AccessFree to download

1 Introduction

2 Background

2.1 Defining Cognitive Performance

2.1.1 Intra-individual Cognitive Variability

2.2 Bias Mitigation in Cognitive Performance Evaluation

2.3 Machine and Deep Learning Models for Cognitive Performance Prediction

2.3.1 Machine Learning Models using Extracted Features

2.3.2 Deep Learning Models for Time Series Analysis

2.4 Machine and Deep Learning Explainability Techniques

2.4.1 Shapley Additive Explanations

2.4.2 Occlusion-Sensitivity Methods

2.5 Diverse Counterfactual Explanations for Explainability

3 State of the Art

3.1 Data Collection Protocols for Cognitive Performance Evaluation

3.2 Experimental Studies on Cognitive Performance

3.2.1 Physical Activity

3.2.2 Sleep

3.2.3 Nutrition

3.2.4 Blood Pressure and Daily Activities

3.3 Machine Learning Models for Cognitive Performance Prediction

3.4 Explainable Artificial Intelligence

3.4.1 Ante-hoc Methods

3.4.2 Post-Hoc Methods

4 Methodology

4.1 Study Design and Data Collection

4.1.1 Research Objective and Participant Selection

4.1.2 Data Collection Framework

4.1.3 Data Collection Period and Experimental Phases

4.1.4 Definition of Cognitive Performance and Bias Mitigation

4.1.5 Ethical Considerations and Informed Consent

4.2 Data Preprocessing and Standardization

4.2.1 Handling Missing Data

4.2.2 Time-Series Normalization

4.3 Machine Learning Approaches for Cognitive Performance Prediction

4.3.1 Machine Learning Models

4.3.2 Deep Learning Models

4.4 Explainability for Machine Learning Models

4.5 Generating Counterfactual Explanations with DiCE

5 Implementation

5.1 Building the Dataset

Towards a Comprehensive Model for Cognitive Performance Prediction: A Machine Learning-Based Approach Tesi di Laurea Magistrale in Biomedical Engineering Author: Laura Ginestretti Student ID: 10682819 Advisor: Prof. Marco D.Santambrogio Co-advisors: Alessandro Verosimile Academic Year: 2024-25 i Abstract As wearable technology advances, the ability to derive meaningful, data-driven insights into cognitive performance and mental well-being has significantly expanded. This thesis presents a pilot study that integrates multimodal data—including nutrition, sleep, physical activity, and real-time physiological monitoring—to track and predict cognitive performance, encompassing memory, reasoning, attention, reading, and learning capabilities. A predictive framework was developed using machine learning (ML) and deep learning (DL) models, with explainable AI (XAI) techniques such as SHAP and Occlusion Sensitivity applied to enhance interpretability. Additionally, counterfactual analysis was leveraged to generate actionable recommendations for cognitive performance optimization, providing insights into personalized lifestyle modifications. By detailing the data collection pipeline, preprocessing strategies, and experimental outcomes, this work represents the first study in the state of the art to correlate lifestyle and physiological data with cognitive performance while accounting for individual cognitive baselines. The findings lay the foundation for real-time, personalized cognitive enhancement strategies based on wearable sensor data. Keywords: cognitive performance, physiological data, lifestyle, machine learning, explainability, counterfactual analysis Abstract in lingua italiana Con l’evoluzione della tecnologia indossabile, la capacità di ottenere informazioni significative e basate sui dati in merito a prestazioni cognitive e benessere mentale è notevolmenteaumentata. Questa tesi presenta pertanto uno studio pilota che integra dati multimodali—inclusi nutrizione, sonno, attività fisica e monitoraggio fisiologico in tempo reale—per tracciare e prevedere le prestazioni cognitive stesse, che comprendonomemoria, ragionamento, attenzione, lettura e capacità di apprendimento. Nel contesto di questo studio pilota è stato sviluppato un framework predittivo basato su modelli di machine learning (ML) e deep learning (DL), con l’applicazione di tecniche di explainable AI (XAI) come SHAP e Occlusion Sensitivity per migliorare l’interpretabilità dei modelli stessi. Inoltre, è stata implementata un’analisi controfattuale per generare raccomandazioni personalizzate volte a ottimizzare le prestazioni cognitive, fornendo suggerimenti pratici su modifiche attuabili nello stile di vita. Attraverso la descrizione della pipeline di raccolta dati, delle strategie di preelaborazione e dei risultati sperimentali, questo lavoro rappresenta il primo studio nello stato dell’arte a correlare dati fisiologici e di stile di vita alle prestazioni cognitive, tenendo conto delle differenze cognitive individuali. I risultati ottenuti gettano quindi le basi per lo sviluppo di strategie personalizzate di miglioramento cognitivo in tempo reale basate sui dati provenienti dai dispositivi indossabili stessi. Parole chiave: prestazione cognitiva, dati fisiologici, stile di vita, machine learning, explainability, analisi controfattuale v Contents Abstract i Abstract in lingua italiana iii Contents v 1 Introduction 1 2 Background 5 2.1 Defining Cognitive Performance ........................ 5 2.1.1 Intra-individual Cognitive Variability ................. 6 2.2 Bias Mitigation in Cognitive Performance Evaluation ............ 7 2.3 Machine and Deep Learning Models for Cognitive Performance Prediction . 8 2.3.1 Machine Learning Models using Extracted Features ......... 9 2.3.2 Deep Learning Models for Time Series Analysis ........... 12 2.4 Machine and Deep Learning Explainability Techniques ........... 18

FAQs

What data sources are used to predict cognitive performance in this thesis?

The study integrates multimodal data, including nutrition, sleep, physical activity, and real-time physiological monitoring.

How does the thesis improve the interpretability of machine learning models?

Explainable AI techniques are applied, including SHAP and occlusion sensitivity methods, to make model outputs more understandable.

What is the role of counterfactual analysis in the proposed framework?

Counterfactual analysis is used to generate actionable recommendations for optimizing cognitive performance through personalized lifestyle modifications.

Aug 3, 2026

Upload by Skyler