道路交通事故至今仍然是全世界人类死亡的主要原因之一，驾驶员本身是导致道路交通事故的主要原因，而驾驶分心是驾驶失误的重要诱因。现有的驾驶分心监测方案在实际应用中存在一定的局限性，本文基于模拟驾驶实验，设计了一种多水平的驾驶员认知分心监测系统，以解决目前认知分心表现一致性差、个体差异性强、难以监测、缺乏多水平的问题。

研究1构建了一个多水平的驾驶认知分心情境，采集了不同年龄段的42名被试在不同水平下的驾驶绩效、眼动数据和认知负荷指标。通过数据处理，获取了不同水平认知分心状态下驾驶员的驾驶绩效、眼动指标及主观负荷的差异情况，其中多项指标会随着认知分心难度增加而显著变化。

研究2根据研究1采集数据，确认了指标数据权重，并藉此构建训练数据集，用于训练精准、高效的多模态、多水平的机器学习认知分心监测系统。研究2基于数据集，完成了机器学习模型训练、测试与对比任务，构建了基于多模态数据的多水平认知分心监测模型。本研究对比多个机器学习模型的监测效果，确定随机森林（Random Forest）算法是适用本研究数据集的最优模型。

在研究方法上，本研究使用了不同难度的认知分心任务来触发不同的驾驶员认知分心。本文提出的这一机器学习驾驶分心监测系统有望提高驾驶安全，降低交通事故发生率，并推动智能交通系统的发展。

Road traffic accidents remain one of the leading causes of death worldwide, with driver error being a primary contributor. Driver distraction is a significant factor in driving errors. Existing driver distraction monitoring solutions have limitations in practical applications. This paper presents a multi–level driver cognitive distraction monitoring system based on simulated driving experiments to address the issues of poor consistency in cognitive distraction manifestations, strong individual differences, difficulty in detection, and lack of multi–level approaches.

Study 1 constructed a multi–level driving cognitive distraction scenario and collected driving performance, eye–tracking data, and cognitive load metrics from 42 participants across different age groups under various levels of distraction. Through data processing, differences in driving performance, eye–tracking indicators, and subjective load under different levels of cognitive distraction were obtained, with several indicators showing significant changes as cognitive distraction difficulty increased.

Study 2 utilized data collected in Study 1 to confirm the weight of indicator data and build a training dataset for a precise, efficient multi–modal, multi–level machine learning cognitive distraction monitoring system. Study 2 completed machine learning model training, testing, and comparison tasks based on the dataset, constructing a multi–level cognitive distraction monitoring model based on multi–modal data. This research compared the detection performance of multiple machine learning models, determining that the Random Forest algorithm is the most suitable model for the collected dataset.

In terms of research methods, this study employed cognitive distraction tasks with varying difficulty levels to trigger the proposed machine learning–based driver distraction monitoring system. This system is expected to enhance driving safety, reduce traffic accident occurrence rates, and promote the development of intelligent transportation systems.