A Connectivity-Aware Graph Neural Network for Real-Time

Abstract

Drowsiness detection plays a critical role in enhancing driver safety and preventing accidents due to fatigue. Our approach integrates the advantages of several advanced machine learning algorithms to improve prediction accuracy and responsiveness. Specifically, we employ a Graph Neural Network (GNN) to model the spatial and temporal dependencies in driver behavior, coupled with a Recurrent Neural Network (RNN) architecture using Gated Recurrent Units (GRU) to capture long-term sequential patterns. Furthermore, the XGBoost algorithm is utilized for feature enhancement, and Random Forest (RF) is used to provide an ensemble learning framework for robust classification. The CAGNN framework is designed to dynamically adjust to real-time changes in connectivity and vehicle environment, ensuring seamless performance even in varying conditions. Experimental results demonstrate that our model significantly outperforms traditional drowsiness detection methods in terms of accuracy, latency, and adaptability to real-world conditions.

Keywords: Drowsiness classification, Graph Neural Network (GNN), Connectivity-aware, Recurrent Neural Networks (RNN), Gated Recurrent Units (GRU), XGBoost, Random Forest (RF), Real-time detection, Fatigue monitoring, Driver safety.