The rapid growth of digital advertising has intensified the need for objective and scalable methods to evaluate consumer responses beyond traditional self-reported measures. Neuromarketing, particularly through electroencephalography (EEG), offers a promising approach to capturing real-time neural signals associated with attention, engagement, and memory formation. However, a central methodological challenge remains: how to transform high-dimensional EEG time series into meaningful and robust representations for predictive modeling. This thesis investigates alternative EEG representation strategies for advertisement recall prediction, comparing handcrafted statistical features with temporally informed embedding approaches (TS2Vec and FEMBA) within a supervised machine learning framework. Using data collected during controlled advertisement exposure sessions, multiple models are trained and evaluated under consistent validation settings to assess predictive performance, generalization capacity, and interpretability. The results provide empirical evidence on the trade-offs between model complexity, temporal modeling capacity, and computational cost. By systematically analyzing these representation strategies, the study contributes to both neuromarketing analytics and applied machine learning, offering practical guidance for designing scalable and interpretable EEG-based advertising evaluation systems.
La rapida crescita della pubblicità digitale ha intensificato la necessità di metodi oggettivi e scalabili per valutare le risposte dei consumatori oltre le tradizionali misure auto-riportate. Il neuromarketing, in particolare attraverso l’elettroencefalografia (EEG), offre un approccio promettente per catturare segnali neurali in tempo reale associati all’attenzione, al coinvolgimento e alla formazione della memoria. Tuttavia, una sfida metodologica centrale rimane: come trasformare serie temporali EEG ad alta dimensionalità in rappresentazioni significative e robuste per la modellazione predittiva. Questa tesi indaga strategie alternative di rappresentazione dell’EEG per la previsione del ricordo pubblicitario, confrontando caratteristiche statistiche artigianali con approcci di embedding informati temporalmente (TS2Vec e FEMBA) all’interno di un framework di machine learning supervisionato. Utilizzando dati raccolti durante sessioni controllate di esposizione pubblicitaria, vengono addestrati e valutati più modelli in condizioni di validazione coerenti per valutare le prestazioni predittive, la capacità di generalizzazione e l’interpretabilità. I risultati forniscono evidenze empiriche sui compromessi tra complessità del modello, capacità di modellazione temporale e costo computazionale. Analizzando sistematicamente queste strategie di rappresentazione, lo studio contribuisce sia all’analitica del neuromarketing sia al machine learning applicato, offrendo indicazioni pratiche per la progettazione di sistemi scalabili e interpretabili di valutazione pubblicitaria basati su EEG.
When does model complexity pay off? A cost-performance comparison of temporal and aggregated EEG representations for advertising recall prediction
Jandaghian, Amirhossein
2025/2026
Abstract
The rapid growth of digital advertising has intensified the need for objective and scalable methods to evaluate consumer responses beyond traditional self-reported measures. Neuromarketing, particularly through electroencephalography (EEG), offers a promising approach to capturing real-time neural signals associated with attention, engagement, and memory formation. However, a central methodological challenge remains: how to transform high-dimensional EEG time series into meaningful and robust representations for predictive modeling. This thesis investigates alternative EEG representation strategies for advertisement recall prediction, comparing handcrafted statistical features with temporally informed embedding approaches (TS2Vec and FEMBA) within a supervised machine learning framework. Using data collected during controlled advertisement exposure sessions, multiple models are trained and evaluated under consistent validation settings to assess predictive performance, generalization capacity, and interpretability. The results provide empirical evidence on the trade-offs between model complexity, temporal modeling capacity, and computational cost. By systematically analyzing these representation strategies, the study contributes to both neuromarketing analytics and applied machine learning, offering practical guidance for designing scalable and interpretable EEG-based advertising evaluation systems.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_03_Jandaghian.pdf
non accessibile
Descrizione: Thesis
Dimensione
2.1 MB
Formato
Adobe PDF
|
2.1 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/10589/252261