Polytetrafluoroethylene (PTFE) film is observed to increase surface roughness during annealing. Longer annealing times leads to greater surface roughness. The coefficient of friction of PTFE film is affected by the shape of microscale sized particles on the film surface. In this study, we investigate the coefficient of friction of PTFE films using a coarse-grained molecular dynamics model based on experimental observations. We observe how the variation in PTFE chain length and film density affect the topography of PTFE films. We also investigate how these properties of PTFE, and the indenter radius affect the coefficient of friction observed during surface scratch. We find that short PTFE chain lengths create a dense film with greater particle spacing, but longer chains form a mesh structure which reduces the density and creates overlapping portions of particles in the film. We develop a convolutional neural network to classify PTFE film surface and predict the coefficient of friction of a modeled film based solely on the equilibrated film topography. The accuracy of the network was seen to increase when the density and images of internal fiber orientation were added as input features. These results indicate that the coefficient of friction of PTFE films in part is governed by the internal structure of the film.