The growing popularity of cycling as a sustainable transportation choice has inten- sified the focus on cyclist safety. Despite this mounting concern, the development of cyclist-side decision support systems aimed at enhancing cyclist safety remains limited. This study aims to bridge this gap by conducting a comprehensive literature review of machine learning-based trajectory prediction methods applied to pedestrians. Build- ing upon this groundwork, the study develops cyclist trajectory prediction models and assesses a cyclist decision support system within a simulated environment. Our study involves a comparative analysis of trajectory prediction models, examining the impact of including social context and scene context in comparison to a baseline model that solely considers the cyclist’s individual dynamics. The evaluation metrics used encompass the Final Displacement Error and Average Displacement Errors. Our analysis reveals a significant improvement in prediction accuracy through the incorporation of social and scene contexts, while social contexts exert a more substantial influence on accurate trajectory prediction. Moreover, the results suggest that when integrating social context, both the proximity of surrounding road users and the density of neighborhoods influ- ence the prediction accuracy. The developed trajectory prediction model is integrated into a simulated environment, coupled with a conflict detection algorithm, to construct a cyclist-decision support system. This integration facilitates the timely provision of collision alerts to cyclists, highlighting potential conflict scenarios. Through simulation- based evaluations, instances of trajectory prediction and conflict detection failures are identified. These findings not only contribute to a comprehensive understanding of system limitations but also pave the way for future research avenues.