ASTIR: Spatio-Temporal Data Mining for Crowd Flow Prediction

Lablack Mourad,Yanming Shen,Heng Qi,Baocai Yin

doi:10.1109/access.2019.2950956

Abstract

The citywide crowd flow prediction is crucial for a city to ensure productivity, safety and management of its citizen. However, the crowd flow may be affected by many factors, such as weather, working times, events, seasons, and so on. In this paper, we proposed Attentive Spatio-Temporal Inception ResNet (ASTIR), which aims to address the difficulty of crowd flow prediction. The ASTIR is based on the Inception-ResNet structure combined with Convolution-LSTM layers and attention module to better capture pattern movement changes. We build our deep neural network framework consisting of four distinct parts, by which we can capture the short-term, long-term and period properties, as well as external factors that can affect crowd flow behaviors. To show the performance of the proposed method, we use the widely applied benchmarks for crowd flow prediction (Taxi Beijing and Bike New York), and obtain notable improvements over the state-of-the-art approaches.

Highlights

Nowadays almost every person is connected via intelligent devices, and those devices have led to an enormous amount of data that is broadcasted in realtime to servers for the purpose of analysis
We focus on a better way to architect a deep neural network model for citywide crowd flow prediction
WORK In this paper, we propose Attentive Spatio-Temporal Inception ResNet (ASTIR), a deep neural network model for predicting citywide crowd flow

Summary

INTRODUCTION

Nowadays almost every person is connected via intelligent devices (smartphones, cars, smartwatches and so on), and those devices have led to an enormous amount of data (personal information, health information, GPS position, etc.) that is broadcasted in realtime to servers for the purpose of analysis. The crowd flow analysis is one of it It is the interpretation of devices signals and GPS position data changes over time, i.e., analyzing the movements of individuals. This analysis will be used to make crowd flow predictions. 2. Temporal dependency: the crowd flow is affected by what happened in the nearby time. We architect ASTIR in a way that it can take into consideration three main properties related to crowd flow data (short-term, long-term, and periodical dependencies). ASTIR applies Convolution-LSTM layers assembled in inception-resnet-v2 way [1], which allows us to have different sizes of filters to capture more accurately the spatio-temporal aspect of crowd flow dataset. ASTIR is extensively evaluated on two popular datasets (Taxi Beijing and Bike NYC) and outperforms the stateof-the-art frameworks by a significant margin

RELATED WORK

CROWD FLOW PREDICTIO

CONCLUSION AND FUTURE WORK