Predicting Cyclist Speed in Urban Contexts: A Neural Network Approach

Bicycle use has become more important today, but more information and planning models are needed to implement bike lanes that encourage cycling. This study aimed to develop a methodology to predict the speed a cyclist can reach in an urban environment and to provide information for planning cycling infrastructure. The methodology consisted of obtaining GPS data on longitude, latitude, elevation, and time from a smartphone of two groups of cyclists to calculate the speeds and slopes through a model based on a recurrent short-term memory (LSTM) type neural network. The model was trained on 70% of the dataset, with the remaining 30% used for validation and varying training epochs (100, 200, 300, and 600). The effectiveness of recurrent neural networks in predicting the speed of a cyclist in an urban environment is shown with determination coefficients from 0.77 to 0.96. Average cyclist speeds ranged from 6.1 to 20.62 km/h. This provides a new methodology that offers valuable information for various applications in urban transportation and bicycle line planning. A limitation can be the variability in GPS device accuracy, which could affect speed measurements and the generalizability of the findings.