Due to the application of various pedestrian-friendly policies, the number of pedestrian accidents in Seoul has decreased. However, the use of smart phones has changed the walking behaviour, and has worsened the pedestrian environment due to disruption of delivery motorcycles or electric scooters. As a result, the decline in pedestrian accidents has abated, whereas the number of minor accidents has increased. A survey of walking behaviour among Seoul citizens indicated that 69.0% of pedestrians use smart phones while walking, and 44.9% and 23.6% of them watch a ‘Video’ or play a ‘Game’ respectively, which leads to vulnerable pedestrian behaviour. In order to reflect this change in pedestrian behaviour, we developed pedestrain big data including smart phone usage, and modelled pedestrian accidents using deep learning techniques. Compared with traditional linear regression, the developed model provides a better estimate of accident rate. We also predicted pedestrian accidents with respect to changes in future pedestrian environment, such as increasing the use of electric scooter. The framework of this study based on deep learning and big-data analysis may be applicable to other areas including prediction of traffic congestion or demand for public transit.