Chloride ions are known to penetrate concrete and cause corrosion of steel reinforcement, leading to a reduction in load-bearing capacity and ultimately resulting in structural failure. Therefore, it is crucial to elucidate the diffusion mechanism of chloride ions. However, existing models have been limited in their ability to accurately predict diffusion behaviour due to their consideration of a single factor. Physical model experiments were conducted to investigate the diffusion mechanism of chloride ions in concrete structures under multi-field coupling conditions. The coupled effects of erosion time, ambient temperature, humidity, coarse aggregate volume fraction, and water-cement ratio on chloride ion diffusion behaviour were analysed. A prediction model for surface chloride concentration (C_s) and chloride diffusion coefficient (D) was subsequently developed using machine learning methods, and the importance of each factor was assessed. The results indicated that the BP neural network model is more effective in simulating the variation patterns of C_s and D. Through a weighting analysis, it is determined that erosion time, ambient temperature, and water-cement ratio has a positive correlation with both C_s and D, whereas the volume fraction of coarse aggregate, identified as one of the most significant influencing factors, exhibited a blocking effect on chloride ion transmission. Additionally, humidity is found to influence the chloride diffusion coefficient and surface chloride concentration, exerting a positive effect on the former and a negative effect on the latter, although its contribution is relatively minor compared to the other factors.