The Chinese Survey Space Telescope (CSST) is expected to characterize the atmospheres of exoplanets, providing new opportunities for exoplanet observations. At present, the atmospheres of exoplanets have mostly been studied by transmission spectroscopy. However, evidence of stellar activity has been found in several observed transmission spectra, and the results show that it is difficult to accurately constrain the stellar contamination. Therefore, how to accurately distinguish and eliminate stellar contamination has become a major challenge in transmission spectroscopy. Consequently, a quantitative assessment of the potential stellar contamination and its distinguishability is required when selecting targets for follow-up transmission spectroscopy observations, which requires large-scale simulations to provide the expected data. Here we use multi-color photometry combined with machine learning to identify potential stellar contamination in the transmission spectra of extrasolar gas planets, and design a filter combination with the highest accuracy for stellar contamination discrimination for the CSST Multi-Channel Imager (MCI), which can provide options for developing observing strategies for follow-up observations of exoplanet atmospheric targets.