A crash feature-based allocation method for boundary crash problem in spatial analysis of bicycle crashes

•Boundary crash problem in spatial analysis is addressed using a data-driven approach.•Individual crash features are considered when allocating the boundary crashes.•Missing data is estimated using augmented masked autoencoder method.•Proposed crash feature-based approach is superior for boundary crash allocation.•Bicycle and traffic flow, gender, household income, land use, and road network characteristics can affect zonal-level bicycle crash frequency.