The invention provides a coding prediction method and system based on artificial intelligence. According to the technical scheme, a multi-stage coding mode is adopted; an adversarial network model isintroduced to filter the input data before and after encoding; input data is further input into a deep learning network to be subjected to coding learning prediction after being subjected to coding prediction in a manner of combining an auto-encoder, and track scene networked modeling and recommendation can be subsequently carried out according to an output result, so that a to-be-decoded sequencewith more output data than that of the original output data is obtained.