Empowering female cyclists’ sustainable mobility: A deep learning approach to predict security perceptions

Cycling is increasingly valued as a sustainable transportation mode, enhancing urban mobility and environmental benefits. However, perceived security (PS) concerns—a key barrier to cycling adoption, especially among women in several countries—highlight a critical social dimension in sustainable urban planning. This study introduces a deep learning (DL)-based predictive model for women cyclists’ PS, addressing a notable research gap. We used a virtual reality (VR) cycling dataset of 210 samples, integrating socio-environmental features such as crowdedness, incivility, lighting conditions, visibility obstructions, and surveillance to assess selected DL architectures. Deep Neural Networks (DNNs), ResNet, Feature Tokenizer Transformer (FTT), and HyperTab—featuring neuron-layer designs, skip connections, and attention mechanisms for tabular data—are examined. The FTT model excels with 79% accuracy, showing robust generalization across key PS categories, and effectively identifying crowdedness, lighting, and visibility as significant determinants of perceived security despite dataset constraints. This research highlights the role, and offers preliminary evidence for the potential of advanced technological solutions in facilitating inclusive urban sustainability. By providing initial practical insights to urban planners, the findings directly contribute to the development of more secure cycling environments, thereby enhancing women’s participation in sustainable transportation systems. Furthermore, this approach establishes a foundation for future interdisciplinary research to leverage larger and more diverse datasets to generalize and refine predictions across different demographic and cultural contexts.