This thesis investigates the perception of academic stress among students at Politecnico di Milano, employing advanced models and Machine Learning algorithms to identify key stressors and potential interventions. The study is motivated by the growing concern over the mental health of university students, with academic stress being a significant con tributing factor. Drawing inspiration from the Perception of Academic Stress Scale (PAS) developed by Bedewy and Gabriel (2015), the research constructs a dataset capturing four main stress factors: Pressures to Perform, Perceptions of Workload and Examinations, Academic Self-Perceptions, and Time Restraints. The methodological approach integrates Item Response Theory (IRT) models to quantify stress and create a generic stress indicator, referred to as the main score. Specifically, the study explores polytomous IRT models including, Rating Scale Model (RSM), Partial Credit Model (PCM), and the Four-Parameter Logistic Regression Model (4PL). The 4PL model emerges as the most effective, capturing the highest proportion of variance in the dataset and providing a robust framework for stress quantification. Additionally, the study employs Random Forest (RF) and Support Vector Machine (SVM) classifiers to identify students at risk of severe psychological conditions, such as suicidal thoughts, self-harm, and panic attacks. The findings highlight the significant impact of academic and non-academic factors on student stress, with economic conditions, exam-related pressures, and social resources be ing key contributors. The IRT-based scores demonstrate strong predictive power in dis tinguishing students with diagnosed mental disorders and those experiencing prolonged psychological distress. Furthermore, the study underscores the importance of early iden tification of at-risk students, offering valuable insights for university policies and support systems. In conclusion, this research provides a comprehensive framework for understanding and mitigating academic stress, leveraging advanced statistical models to offer actionable rec ommendations for improving student well-being.
Questa tesi indaga la percezione dello stress accademico tra gli studenti del Politecnico di Milano, utilizzando modelli avanzati e algoritmi di Machine Learning per identificare i principali fattori di stress e potenziali interventi. Lo studio è motivato dalla crescente preoccupazione per la salute mentale degli studenti universitari, con lo stress accademico che rappresenta un fattore significativo. Ispirandosi alla Perception of Academic Stress Scale (PAS) sviluppata da Bedewy e Gabriel (2015), la ricerca costruisce un dataset che cattura quattro principali fattori di stress: Pressioni per la performance, Percezione del carico di lavoro e degli esami, Autopercezione accademica e Vincoli di tempo. L’approccio metodologico integra modelli di Item Response Theory (IRT) per quantificare lo stress e creare un indicatore generico di stress, denominato main score. In particolare, lo studio esplora modelli IRT politomici tra cui il Rating Scale Model (RSM), il Partial Credit Model (PCM) e il modello di regressione logistica a quattro parametri (4PL). Il modello 4PL si rivela il più efficace, catturando la proporzione più alta di varianza nel dataset e fornendo un quadro robusto per la quantificazione dello stress. Inoltre, lo studio utilizza classificatori come Random Forest (RF) e Support Vector Machine (SVM) per identificare gli studenti a rischio di condizioni psicologiche gravi, come pensieri suicidi, autolesionismo e attacchi di panico. I risultati evidenziano l’impatto significativo di fattori accademici e non accademici sullo stress degli studenti, con condizioni economiche, pressioni legate agli esami e risorse sociali che emergono come contributori chiave. I punteggi basati sull’IRT dimostrano un forte potere predittivo nel distinguere studenti con disturbi mentali diagnosticati e quelli che sperimentano disagio psicologico prolungato. Inoltre, lo studio sottolinea l’importanza dell’identificazione precoce degli studenti a rischio, offrendo spunti preziosi per le politiche universitarie e i sistemi di supporto. In conclusione, questa ricerca fornisce un quadro completo per comprendere e mitigare lo stress accademico, sfruttando modelli statistici avanzati per offrire raccomandazioni pratiche per migliorare il benessere degli studenti.
A statistical analysis of the academic psychological well-being
MARINO, FILIPPO
2023/2024
Abstract
This thesis investigates the perception of academic stress among students at Politecnico di Milano, employing advanced models and Machine Learning algorithms to identify key stressors and potential interventions. The study is motivated by the growing concern over the mental health of university students, with academic stress being a significant con tributing factor. Drawing inspiration from the Perception of Academic Stress Scale (PAS) developed by Bedewy and Gabriel (2015), the research constructs a dataset capturing four main stress factors: Pressures to Perform, Perceptions of Workload and Examinations, Academic Self-Perceptions, and Time Restraints. The methodological approach integrates Item Response Theory (IRT) models to quantify stress and create a generic stress indicator, referred to as the main score. Specifically, the study explores polytomous IRT models including, Rating Scale Model (RSM), Partial Credit Model (PCM), and the Four-Parameter Logistic Regression Model (4PL). The 4PL model emerges as the most effective, capturing the highest proportion of variance in the dataset and providing a robust framework for stress quantification. Additionally, the study employs Random Forest (RF) and Support Vector Machine (SVM) classifiers to identify students at risk of severe psychological conditions, such as suicidal thoughts, self-harm, and panic attacks. The findings highlight the significant impact of academic and non-academic factors on student stress, with economic conditions, exam-related pressures, and social resources be ing key contributors. The IRT-based scores demonstrate strong predictive power in dis tinguishing students with diagnosed mental disorders and those experiencing prolonged psychological distress. Furthermore, the study underscores the importance of early iden tification of at-risk students, offering valuable insights for university policies and support systems. In conclusion, this research provides a comprehensive framework for understanding and mitigating academic stress, leveraging advanced statistical models to offer actionable rec ommendations for improving student well-being.File | Dimensione | Formato | |
---|---|---|---|
2025_04_Marino_Executive Summary.pdf
non accessibile
Descrizione: Testo Executive summary
Dimensione
686.54 kB
Formato
Adobe PDF
|
686.54 kB | Adobe PDF | Visualizza/Apri |
2025_04_Marino_Tesi.pdf
non accessibile
Descrizione: Testo Tesi
Dimensione
5.25 MB
Formato
Adobe PDF
|
5.25 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/235503