Shockley's lucky electron model has been widely used for modeling impact ionization. Various nonlocal generalizations of this model have been proposed in order to model impact ionization more accurately for strongly inhomogeneous material and field conditions which exist, e.g., in n-channel metal-oxide-semiconductor field-effect transistors (MOSFETs). Recently it has been shown that the spatial distribution of impact ionization events resulting from the nonlocal variant is in close agreement with corresponding results derived from an advanced Monte Carlo model. However, the former lucky electron model fails to closely reflect the underlying microscopic processes and is based on the restrictive assumption that the electron trajectories are electrostatic field lines. In order to incorporate more microscopic features, an improved lucky electron modeling approach including ballistic and diffusive transport effects is proposed, which yields comparably favorable results if compared to Monte Carlo data.