With the development of autonomous navigation for unmanned ships, identifying ship collision avoidance behavior has become a key factor in their independent decision-making. To address the inefficiency and misjudgment issues of existing ship trajectory recognition algorithms, this paper proposes a data mining model based on the steering point of a sliding window for ship collision avoidance. When the model identifies a ship's steering point, it first evaluates the change characteristics of the heading at adjacent time points in the ship's Automatic Identification System (AIS) data using a fixed sliding window. Then, the slope change of the trajectory points at adjacent moments is calculated for verification, and the earliest turning point of the heading change within the window is marked. Finally, a variable sliding window is used to maintain the heading change and error parameters during the trajectory change process, determining whether the steering point is a collision-avoidance steering point. The model is experimentally compared with the Douglas-Peucker (DP) algorithm. The results show that the model can effectively identify whether a ship's steering is collision avoidance behavior, resolve the issue of the DP algorithm misjudging steering points due to data fluctuations, and extract the earliest steering point during the ship collision avoidance process to assist in collision avoidance decision-making. This model can be applied to the research and development of intelligent collision avoidance decision-making systems, ensuring the safety of ship navigation.