Bayesian network approach to building an affective module for a driver behavioural model

This paper focuses on the affective component of a driver behavioural model (DBM). This component specifically models some drivers’ mental states such as mental load and active fatigue, which may affect driving performance. We have used Bayesian networks (BNs) to explore the dependencies between various relevant random variables and assess the probability that a driver is in a particular mental state based on their physiological and demographic conditions. Through this approach, our goal is to improve our understanding of driver behaviour in dynamic environments, with potential applications in traffic safety and autonomous vehicle technologies.

💡 Research Summary

This paper presents a Bayesian network (BN)–based affective module for a driver behavioural model (DBM) that aims to quantify drivers’ mental load and active fatigue, two cognitive states known to impair driving performance. The work is part of the European BER‑THA project, which seeks to create a comprehensive, modular DBM integrating cognitive, affective, perceptual, and contextual factors within a unified probabilistic framework.

Data collection and variables
The authors collected data from 17 participants in a driving‑simulator experiment conducted at the Instituto de Biomecánica de Valencia (IBV). Each subject provided demographic information (sex, age, body‑mass index) and physiological recordings (standard deviation of RR intervals – Y_SRT, standard deviation of successive RR differences – Y_SDD, mean heart rate – Y_MHR, low‑to‑high frequency power ratio – Y_RLH, and mean breaths per minute – Y_MNB). After each cognitive block, participants completed validated questionnaires measuring mental load (Y_ML) and active fatigue (Y_AF). The questionnaire responses were binarised (1 = state present, 0 = absent).

Bayesian network structure
A directed acyclic graph (DAG) was designed to capture the causal relationships among the variables. The two mental‑state nodes (Y_ML and Y_AF) directly influence three physiological nodes (Y_MHR, Y_SDD, Y_RLH). Y_SDD also affects Y_MHR, and together Y_SDD, Y_MHR, and Y_RLH determine Y_SRT, which finally serves as a parent of Y_MNB. This hierarchy reflects the authors’ hypothesis that mental states first affect cardiac dynamics, which then propagate to respiratory patterns.

Statistical modelling
Both mental‑state variables are modelled as Bernoulli outcomes using Bayesian logistic regression (BLR). Each continuous physiological variable is modelled with a Bayesian regression model (BRM) assuming a normal likelihood whose mean is a linear function of the three demographic covariates (sex, age, BMI) and the relevant parent nodes. Non‑informative priors are assigned: regression coefficients follow N(0, 25) and the residual standard deviations follow Uniform(0, 30). This prior choice places the inferential weight on the observed data.

Inference procedure
Posterior distributions of all parameters (θ) are approximated via Markov chain Monte Carlo (MCMC) sampling using JAGS. The resulting posterior samples are used to construct the joint posterior predictive distribution f(y | D). From this distribution the authors extract conditional probabilities of interest, such as the probability of experiencing mental load given a specific combination of gender, age, BMI, Y_SRT, and Y_MNB.

Key findings
Figure 3 (described in the text) illustrates average posterior probabilities for the four joint states of active fatigue and mental load across different driver ages, with gender and BMI held constant. The authors report that younger drivers exhibit higher probabilities of both mental load and fatigue, whereas age appears to have a protective effect. Gender, however, does not show a statistically meaningful impact on the probabilities. An example calculation shows that a 20‑year‑old female with BMI 22 and a breathing rate of 20 breaths per minute has an estimated 0.72 probability of being in a mental‑load state.

Discussion and limitations
The study demonstrates that BN can provide real‑time, probabilistic assessments of driver mental states based on readily available physiological signals. Such assessments could be integrated into in‑vehicle safety systems that issue fatigue warnings or suggest breaks. Nevertheless, the authors acknowledge several limitations: the sample size (n = 17) is small, limiting statistical power and generalisability; only a limited set of physiological markers were used; and the model omits other potentially informative modalities such as facial expression, eye‑tracking, or vehicle control signals.

Future work
The authors propose extending the BN with additional sensors (e.g., video‑based facial affect detection) and testing the model on larger, more diverse datasets. They also suggest exploring computationally efficient inference methods (e.g., variational inference) to enable real‑time deployment in commercial vehicles.

Conclusion
By integrating Bayesian network modelling with experimental physiological data, the paper provides a proof‑of‑concept that driver mental load and active fatigue can be quantified probabilistically. This approach offers a solid foundation for developing driver‑monitoring systems that enhance road safety and support the human‑machine interaction challenges of connected, cooperative, and automated mobility (CCAM).