Comparing Deep Learning and Statistical Methods in Forecasting Crowd Distribution from Aggregated Mobile Phone Data

Alket Cecaj,Marco Lippi,Marco Mamei,Franco Zambonelli

doi:10.3390/app10186580

Alket Cecaj, Marco Lippi + Show 2 more

Open Access

https://doi.org/10.3390/app10186580

Copy DOI

Journal: Applied sciences	Publication Date: Sep 21, 2020
Citations: 21	License type: CC BY 4.0

Affiliation: University of Modena and Reggio Emilia

Abstract

Accurately forecasting how crowds of people are distributed in urban areas during daily activities is of key importance for the smart city vision and related applications. In this work we forecast the crowd density and distribution in an urban area by analyzing an aggregated mobile phone dataset. By comparing the forecasting performance of statistical and deep learning methods on the aggregated mobile data we show that each class of methods has its advantages and disadvantages depending on the forecasting scenario. However, for our time-series forecasting problem, deep learning methods are preferable when it comes to simplicity and immediacy of use, since they do not require a time-consuming model selection for each different cell. Deep learning approaches are also appropriate when aiming to reduce the maximum forecasting error. Statistical methods instead show their superiority in providing more precise forecasting results, but they require data domain knowledge and computationally expensive techniques in order to select the best parameters.

Highlights

Being able to model and eventually forecast population density and crowd distributions in a smart city scenario is mainly about studying and analyzing data in space and time domain
We describe three different forecasting methods based on deep learning neural networks that we will use in our forecasting scenario
The main conclusions that we reach with this research work are that deep learning methods are easier to apply on time series forecasting tasks

Summary

Introduction

Being able to model and eventually forecast population density and crowd distributions in a smart city scenario is mainly about studying and analyzing data in space and time domain. There are different scientific works that have been studying mobile phone data in depth revealing the different patterns that emerge [1,2,3,4]. Another line of research for aggregated mobile phone data such as in [5] deals with modelling them as a time series, so as to explore their predictability by using different forecasting methods. We aim to underline the strengths and weaknesses of each category of methods and, most importantly, to investigate which one might be the most appropriate for a specific forecasting scenario

Objectives

Methods

Findings

Conclusion