Revisión sistemática: El uso de inteligencia artificial para la detección de fatiga en conductores para la prevención de accidentes de tránsito

Abstract

Fatigue while driving increases the risk of road accidents due to the decline in the driver's cognitive and motor skills. The objective was to systematically analyze the use of artificial intelligence in detecting fatigue in drivers, evaluating its effectiveness, the technologies used, and its potential contribution to reducing traffic accidents. The study is justified because it allows the identification of the most effective artificial intelligence tools and ways of implementation. The methodology was based on PRISMA, selecting 57 articles that met the inclusion criteria. The results identified five main categories as electroencephalogram-based systems with machine learning, computer vision, multimodal solutions, optimization for embedded devices, and the use of complementary physiological signals. Deep learning architectures achieved accuracies above 90%, with convolutional networks, LSTM, and transformers standing out. Classic algorithms, such as Support Vector Machine, achieved up to 91.8% accuracy, while Random Forest and decision trees exceeded 89%; suggesting that combining electroencephalographic signals from a few electrodes with multimodal approaches and explainable models optimizes early detection. It is concluded that mass adoption requires non-invasive systems, lightweight models compatible with the Internet of Things, and continuous validation in natural driving contexts. Keywords: Fatigue, vehicular drowsiness, AI, accident prevention