If you use this dataset, please cite the following paper: John Violos, Theodoros Theodoropoulos, Angelos-Christos Maroudis, Aris Leivadeas, and Konstantinos. Tserpes, “Self-Attention based Encoder-Decoder for Multistep Human Density Prediction,” Elsevier in Journal of Urban Mobility, vol. 2, p. 100022, Dec. 2022, doi: 10.1016/j.urbmob.2022.100022. The simulation took place from 2021-07-16 to 2021-07-23 in the New York Central park with SUMO (Simulation of Urban MObility) [2]. The New York Central park is divided into six Regions Of Interest (ROI) and every time step we estimated the number of people gathered into every ROI as we can see in Fig. 1. We used a time step of 5 mins. In the file Final_regions_timestep.csv the 1st column represents the number of the step and the following six steps represent the six ROIs. In the files 5min_timestep_Day_i.csv (with i from 1 to 7) we can see the SUMO generated data for the i day of the simulation. In every row we can see the id of the pedestrian, the date, the timestamp, the latitude, the longitude. The sixth and the seventh columns are not taken into consideration in our experiments. The sixth column is a constant of SUMO and the seventh is the velocity of the pedestrians. More info can be found in the Dataset Description