The analysis of human behavior is a complex and evolving research field that enables a holistic understanding of daily activities and habits. Existing approaches focus on classifying individual activities, struggling to provide a clear and structured representation of behavioral patterns. Differently, process mining offers a way to create structured and easily interpretable representations of complex behavioral sequences. In this work, we propose a novel approach that leverages a Long Short-Term Memory network and an attention mechanism to abstract high-level activities from fine-grained sensor data in smart environments. The attention layer enhances explainability by clarifying the inference process during activity recognition. These activities are then transformed into structured event logs for process mining analysis. We evaluate our approach using the CASAS Cairo dataset, demonstrating its effectiveness in extracting meaningful activity sequences and supporting structured behavior analysis.