Supervised Contrastive Representation Learning and Single-Stage Mixture Density Fusion for Uncertainty-Aware Classification of Driving Behavior

Abstract

This work presents a unified probabilistic framework for driving-behavior classification using multimodal physiological sensing. Real-world driver-monitoring systems often rely on deterministic fusion networks that struggle when faced with noise, modality corruption, or ambiguous physiological patterns. To address these limitations, we propose a two-stage learning pipeline that combines Supervised Contrastive Representation Learning (SCL) with a Single-Stage Mixture Density Fusion (MDN-Fusion) module.