Process mining is a technique used for analyzing and optimizing business processes through the extraction of process models from event logs. However, the quality of these models is heavily dependent on the quality of the event logs. Anomalous activities, characterized by irregularly ordered or omitted events in different cases, pose a significant challenge to the continuity of the process. To address this challenge, we introduce a window-based online approach. Our approach involves encoding multiple dimensions of real-time event data to detect anomalous activities using unsupervised learning models. This paper presents an innovative solution aimed at enhancing the quality of discovered process models, ultimately improving the efficiency and effectiveness of process mining.