The bridge between Data Science and Process Science, the so-called Process Mining, allow to explore process data in the form of event-logs: sequence of events grouped by case identifier (to identify process instances), sorted by timestamps and assuming the meaning by the activity label associated with each event. A sub domain of Process Mining, called Process Discovery, aims to analyse sequence of events to highlight relationships between events to build up a Process Model. In some event’s sequence related to the same process instances some activities are duplicated during the process execution bringing noise and needing more computational power to discover the Process Model. In this work, it is proposed a time-window activity aggregation technique to reduce the number of duplicated activities in order to discover Process Model with lower levels of noise and simpler using a simpler event-log. The analysis involved the use of performance metrics as Fitness and Precision based on token replay, showing that using a time-window of 60 s allow to obtain a Process Model with same performances but with low levels of noise and computational needs.