<p>In the context of transit-oriented development (TOD), the comprehensive commercial development around metro stations has become a new trend. However, increased commercial attractions can alter passenger behavior, intensifying space congestion and flow conflicts within metro stations. Consequently, commercial service areas can become critical risk zones. This study explored a recognition and evaluation method for consumption behavior based on passenger trajectory data from TOD metro stations. First, the characteristics of both transfer and consumption behaviors are detailed, distinguishing between strong- and weak-purpose consumption. A dynamic model of the “transfer–consumption–transfer” behavior process is also developed. Second, a machine learning-based method for recognizing and evaluating consumption behavior, grounded in passenger trajectory analysis, is proposed. This method employs machine learning method to detect and track passenger movements. Simultaneously, a coordinate transformation model is constructed to correct data deviations from the pixel-to-world coordinate system. Several analytical indicators, including deviation angles, distances, and conflict points, are introduced to mathematically describe the consumption behavior characteristics at TOD metro stations. Finally, the proposed machine learning-based method is applied to a comprehensive metro station in Shanghai, China. The experimental results validate the effectiveness of the proposed method.</p>