Hybrid Quantum-Classical Neural Network for Incident Detection

Abstract

The effectiveness and dependability of real-time incident detection models directly impact the safety and operational conditions of the affected traffic routes. Recent advancements in cloud-based quantum computing infrastructure and developments in noisy intermediate-scale quantum devices have ushered in a new era of quantum-enhanced algorithms that can be utilized to enhance the accuracy of real-time incident detection. In this study, a combination of classical and quantum machine learning (ML) models is developed to identify incidents using connected vehicle (CV) data. The performance of the hybrid classical-quantum machine learning model in incident detection is compared to baseline classical ML models. The framework is evaluated using data from a microsimulation tool that simulates different incident scenarios. Results show that a hybrid neural network with a 4-qubit quantum layer outperforms all other baseline models considered in this study, even when training data is scarce. Three datasets made are DS-1 with enough training data, DS-2, and DS-3 without enough training data. The hybrid approach produces recall rates of 98.9%, 98.3%, and 96.6% for DS-1, DS-2, and DS-3, respectively. For DS-2 and DS-3, the hybrid model outperforms the classical models on average by 1.9% and 7.8%, respectively, for F2-score, a metric for the model’s ability to correctly recognize occurrences. These results indicate that in realistic scenarios with limited connected vehicles at certain times on specific roadways, the hybrid ML model performs better than classical models. As quantum computing infrastructure continues to improve, quantum ML models could offer a promising alternative for ML applications related to connected vehicles, especially when the available data is insufficient.