Mountain roads are the channels for survival in mountainous areas and communication with cities. They are an important support for promoting economic development in mountainous areas. Therefore, it is necessary to establish a prediction model for the accessibility capacity of mountain roads. Combination model refers to the combination of multiple network models to improve prediction accuracy by extracting the advantages of different models. In order to predict the accessibility of mountainous highways, this paper proposes a CNN LightGBM prediction model that combines CNN and LightGBM. This model first utilizes the CNN model to explore the correlation between factors that affect the accessibility of mountainous highways, and constructs an evaluation index system for the accessibility of mountainous highways. Then, this study inputs the extracted feature vectors into the LightGBM model to achieve accessibility prediction. The experimental results indicate that this model has better predictive performance than previous models. This research can help traffic planners make more informed decisions in determining the priority of maintenance and upgrading projects by more accurate assessment of road conditions. And this method can be applied to other regions and road types, which provides valuable insights into the potential of in-depth learning methods in transport infrastructure management.