[论文解读] Dynamically Weighted Ensemble-based Prediction System for Adaptively Modeling Driver Reaction Time

Chun‐Hsiang Chuang, Zehong Cao+4 more|arXiv (Cornell University)|Sep 18, 2018

Sleep and Work-Related Fatigue参考文献 52被引用 11

一句话总结

本文提出了一种动态加权集成系统，通过结合多个在EEG数据上训练的子模型，自适应地建模驾驶员反应时间（RT），这些子模型具有相似的EEG-RT关系。该系统使用10通道的实时EEGα波段相干性和30通道的θ波段功率作为自适应权重，显著提高了传统方法在RT预测精度方面的表现，展现出对个体差异和时间变化的脑-行为关系的鲁棒性。

ABSTRACT

Predicting a driver's cognitive state, or more specifically, modeling a driver's reaction time (RT) in response to the appearance of a potential hazard warrants urgent research. In the last two decades, the electric field that is generated by the activities in the brain, monitored by an electroencephalogram (EEG), has been proven to be a robust physiological indicator of human behavior. However, mapping the human brain can be extremely challenging, especially owing to the variability in human beings over time, both within and among individuals. Factors such as fatigue, inattention and stress can induce homeostatic changes in the brain, which affect the observed relationship between brain dynamics and behavioral performance, and thus make the existing systems for predicting RT difficult to generalize. To solve this problem, an ensemble-based weighted prediction system is presented herein. This system comprises a set of prediction submodels that are individually trained using groups of data with similar EEG-RT relationships. To obtain a final prediction, the prediction outcomes of the sub-models are then multiplied by weights that are derived from the EEG alpha coherences of 10 channels plus theta band powers of 30 channels, whose changes were found to be indicators of variations in the EEG-RT relationship. The results thus obtained reveal that the proposed system with a time-varying adaptive weighting mechanism significantly outperforms the conventional system in modeling a driver's RT. The adaptive design of the proposed system demonstrates its feasibility in coping with the variability in the brain-behavior relationship. In this contribution surprisingly simple EEG-based adaptive methods are used in combination with an ensemble scheme to significantly increase system performance.

研究动机与目标

• 解决由于脑-行为关系的个体差异和时间差异导致的驾驶员反应时间（RT）建模挑战。
• 提升RT预测系统在真实驾驶环境中的泛化能力和准确性。
• 开发一种自适应框架，根据实时EEG认知状态指标动态调整预测权重。
• 验证EEG衍生特征——特别是α波段相干性和θ波段功率——作为集成模型动态加权信号的有效性。
• 证明简单、自适应的集成方法在RT预测中优于静态或非自适应预测系统。

提出的方法

• 该系统采用多个子模型的集成，每个子模型均在具有相似EEG-RT关系的数据段上进行训练。
• 使用10个通道的EEG α波段相干性和30个通道的θ波段功率动态计算预测权重。
• 加权机制可实时调整，以反映因疲劳、注意力不集中或压力导致的EEG-RT关系变化。
• 通过这些时变权重组合子模型的输出，生成最终的RT预测结果。
• 用于加权的EEG特征基于其对脑-行为耦合变化的敏感性进行选择。
• 该系统在模拟驾驶任务中采集的EEG数据上进行训练和验证，任务中包含危险检测事件。

实验结果

研究问题

• RQ1 动态加权集成系统是否能在认知状态变化的情况下提升驾驶员反应时间预测的准确性？
• RQ2 哪些EEG特征最能指示因疲劳或注意力不集中导致的EEG-RT关系变化？
• RQ3 自适应加权机制如何提升系统在个体之间和随时间推移的泛化能力？
• RQ4 像α波段相干性和θ波段功率这样的简单EEG生物标志物能否有效作为动态加权信号？
• RQ5 所提出的系统是否在RT预测中优于传统的静态或非自适应集成模型？

主要发现

• 所提出的动态加权集成系统在建模驾驶员反应时间方面显著优于传统预测系统。
• 使用EEG α波段相干性和θ波段功率作为动态权重，通过适应认知状态变化，提升了预测精度。
• 该系统在不同个体和长时间跨度内均表现出稳健性能，表明其具备强大的泛化能力。
• 自适应加权机制有效捕捉了因疲劳和压力导致的脑-行为关系的变异。
• 结果证实，当整合到自适应集成框架中时，简单的EEG特征可发挥高度有效性。
• 在真实EEG数据集上的验证表明，该系统在性能上优于非自适应或固定权重的集成模型。