Quantum Machine Learning Models Pedestrian Stress for Safer Roads

Research demonstrates that a quantum neural network (QNN), utilising an eight-qubit ZZ feature map, achieves 55% test accuracy in classifying pedestrian stress levels from skin conductance responses in a virtual reality environment. This surpasses both a quantum support vector machine and classical models, indicating improved classification performance. Skin conductance response, also known as galvanic skin response, measures changes in sweat gland activity, reflecting emotional arousal.

The increasing complexity of urban environments demands sophisticated analytical tools to understand human behaviour within them, particularly concerning pedestrian safety. Researchers are now investigating whether quantum machine learning offers advantages over classical approaches in modelling nuanced physiological responses to perceived threat. Bara Rababah, Bilal Farooq, and colleagues from the Laboratory of Innovations in Transportation at Toronto Metropolitan University detail their work in “Quantum Machine Learning in Transportation: A Case Study of Pedestrian Stress Modelling”, where they explore the application of quantum algorithms to analyse skin conductance response – a physiological indicator of stress – gathered during a virtual reality road-crossing experiment. Their study compares the performance of a quantum support vector machine and a quantum neural network with classical machine learning models, assessing their ability to classify pedestrian stress levels accurately.

This research investigates the application of quantum machine learning (QML) to interpret pedestrian stress, utilising electrodermal activity (EDA) data gathered during a virtual reality (VR) road-crossing simulation. Electrodermal activity, also known as skin conductance, reflects changes in sweat gland activity and is a physiological indicator of emotional arousal and stress. Researchers explore whether QML algorithms offer advantages in modelling complex physiological responses within intelligent transportation systems (ITS), comparing the performance of a Quantum Support Vector Machine (QSVM) and a Quantum Neural Network (QNN) against their classical counterparts. The study addresses a critical need for improved methods to analyse high-dimensional data, crucial for understanding pedestrian behaviour and enhancing urban planning initiatives.

The research team meticulously collected skin conductance response (SCR) measurements, alongside detailed features describing response amplitude and elapsed time, categorising these into amplitude-based classes for comprehensive analysis. Researchers implemented a QSVM, employing an eight-qubit ZZ feature map, which initially demonstrated strong training accuracy; however, it subsequently suffered from significant overfitting, resulting in a low test accuracy of 45%. Overfitting occurs when a model learns the training data too well, including its noise and irregularities, and consequently performs poorly on unseen data. This outcome limits the reliability of the QSVM as a classification tool.

In contrast, the QNN achieved superior performance. Researchers found that the QNN’s ability to capture non-linear relationships in the data is crucial for accurately classifying different stress levels, and that the Tree Tensor Network ansatz effectively reduces the number of parameters required to train the model, preventing overfitting and improving generalisation performance. A tensor network is a mathematical construct used to represent multi-dimensional arrays. In this context, it helps to efficiently represent and manipulate the quantum states within the neural network.

Researchers highlight the potential of QML to address the challenges of high-dimensional data analysis in ITS, noting that the ability to process and extract meaningful information from complex physiological signals efficiently can lead to significant improvements in pedestrian safety and urban planning. The team envisions a future where QML-powered systems can proactively identify and mitigate stress factors in urban environments, creating more comfortable and enjoyable experiences for pedestrians, and potentially personalise urban environments based on individual pedestrian stress responses.

Researchers meticulously analysed the results, identifying the key factors that contribute to the superior performance of the QNN model. The ability of the QNN to model complex interactions within the EDA data, coupled with the efficient parameterisation provided by the Tree Tensor Network, appears to be central to its success. Researchers also highlight the potential of QML to revolutionise the field of ITS, noting that the ability to process and analyse complex data streams efficiently can lead to significant improvements in traffic management, pedestrian safety, and urban planning. The team envisions a future where QML-powered systems can proactively identify and mitigate potential hazards in urban environments, creating safer and more enjoyable experiences for all, and optimise traffic flow and reduce congestion.

Researchers acknowledge the limitations of the current study, including the small sample size and the limited complexity of the quantum circuits. The team plans to address these limitations in future work by collecting data from a larger and more diverse population and by exploring more sophisticated quantum algorithms, and investigating the potential of combining QML with other machine learning techniques to further improve the accuracy and robustness of the models.