This research project will explore psychological risk and protective trajectories during pregnancy through an integrated methodology combining relational dimensions and predictive analysis. The study will adopt a longitudinal, multimodal design involving approximately 1,000 pregnant women, with a subsample (n≈200) selected for vocal data collection.

The project will unfold in interconnected phases: first, Cluster Analysis and Latent Class Analysis will identify distinct relational and psychopathological profiles based on emotional distress, parental representations, couple relationship quality, and prenatal attachment. These profiles will then serve as input features for supervised machine learning models (Random Forest, Gradient Boosting, SVM) to predict postpartum psychopathological vulnerability and mother-infant bonding quality. Concurrently, maternal voice analysis—examining prosodic features (intonation, rhythm, pauses) and semantic content (emotional representations, narrative quality)—will explore vocal markers as early indicators of psychological vulnerability and prenatal bonding difficulties.

By integrating clinical, narrative, and computational approaches, this project aims to develop sensitive tools for early identification and prevention in perinatal care, offering an innovative, scalable response to contemporary maternal-infant health challenges while harmonizing technological innovation with a human-centered approach.