Detecting anomalies in spatio-temporal flow data by constructing dynamic neighbourhoods

Highlights

• The proposed algorithm is designed for anomaly detection in spatio-temporal flow data.

• The spatio-temporal neighbourhoods are constructed based on the modelling of dynamic flow.

• The proposed algorithm can accurately detect global and local spatio-temporal flow anomalies.

Abstract

In massive spatio-temporal datasets, anomalies that deviate from the global or local distributions are not just useless noise but possibly imply significant changes, surprising patterns, and meaningful insights, and because of this, detection of spatio-temporal anomalies has become an important research hotspot in spatio-temporal data mining. For spatio-temporal flow data (e.g., traffic flow data), the existing anomaly detection methods cannot handle the embedded dynamic characteristic. Therefore, this paper proposes the approach of constructing dynamic neighbourhoods to detect the anomalies in spatio-temporal flow data (called spatio-temporal flow anomalies). In this approach, the dynamic spatio-temporal flow is first modelled based on the real-time attribute values of the flow data, e.g., the velocity of vehicles. The dynamic neighbourhoods are then constructed by considering attribute similarity in the spatio-temporal flow. On this basis, global and local anomalies are detected by employing the idea of the G^⁎ statistic and the problem of multiple hypothesis testing is further addressed to control the false discovery rate. The effectiveness and practicality of our proposed approach are demonstrated through comparative experiments on traffic flow data from the central road network of central London for both weekdays and weekends.

Introduction

The wide usage of geo-location sensors and network connectivity makes it easier to capture enormous amounts of spatio-temporal data spanning certain spatial regions over a period of time. Spatio-temporal anomalies are a collection of records that significantly deviate from the global or local distributions with the consideration of non-spatial attributes (Shekhar, Lu, & Zhang, 2001). In most geographical fields, spatio-temporal anomalies are not just useless noises but imply significant changes, surprising patterns, and meaningful insights. Therefore, the detection of spatio-temporal anomalies has become a research hotspot in the field of spatio-temporal data mining (Miller and Han, 2009, Shekhar et al., 2009, Tan et al., 2006) and has received more attention in the detection of disease and crime hotspots (Kulldorff et al., 2005, Delmelle et al., 2014, Brunsdon et al., 2007, Nakaya and Yano, 2010, Shiode and Shiode, 2013, Cheng and Adepejue, 2013), the discovery of climate change (Sun et al., 2005, Barua and Alhajj, 2007, Wu et al., 2010, Liu et al., 2011, Telang et al., 2014), the monitoring of environment change (Birant and Kut, 2006, Cheng and Li, 2006), the extraction of anomalous trajectories (Ge et al., 2010, Lee et al., 2008, Zhang et al., 2011) and the detection of traffic congestion (Li et al., 2007, Liu et al., 2011, Pang et al., 2011, Chawla et al., 2012, Pan et al., 2013).

The types of spatio-temporal data can vary with different applications. In general, the detection methods of existing spatio-temporal anomalies are mostly designed for spatio-temporal sequence data, e.g., climate spatio-temporal sequences. Spatial and non-spatial attributes are both embedded in the spatio-temporal sequences. Specifically, spatial attributes determine the geographical locations of entities by X-Y coordinates or latitude-longitude, while non-spatial attributes are described by time series. Taking Fig. 1(a) as an example, the entity P₁ can be denoted as P₁ = (x₁, y₁, P₁ · nsa_t1, P₁ · nsa_t2, …), where x₁, y₁ and P₁ · nsa_t1 , P₁ · nsa_t2 , … represent the spatial location and non-spatial attribute values of P₁, respectively. In the real world, there exists a kind of special spatio-temporal sequence data that has the characteristics of directionality and being dynamic, e.g., traffic flow data from a road network (also called ‘spatio-temporal flow data’ in this paper). In traffic flow data, the vehicles keep running on the roads, as shown in Fig. 1(b), and the traffic flow in the upstream direction can interact with that in the downstream direction. This paper specifically focuses on spatio-temporal flow data and presents the approach of constructing dynamic neighbourhoods to detect spatio-temporal flow anomalies. The major contributions are as follows:

• Constructing dynamic spatio-temporal neighbourhoods.

• Detecting various anomalies by integrating dynamic spatio-temporal neighbourhoods.

The rest of this paper is organized as follows. Section 2 reviews the related work about the detection of spatio-temporal anomalies and proposes our strategy for detecting spatio-temporal flow anomalies. In Section 3, the proposed method for detecting spatio-temporal flow anomalies is fully elaborated. In Section 4, extensive experiments on real-life data are analysed to demonstrate the effectiveness and practicability of the proposed method. The interesting findings are summarized and future research works are highlighted in Section 5.