Predictive complex event processing based on evolving Bayesian networks

doi:10.1016/j.patrec.2017.05.008

Pattern Recognition Letters

Volume 105, 1 April 2018, Pages 207-216

https://doi.org/10.1016/j.patrec.2017.05.008 Get rights and content

Highlights

•
Bayesian network model uses Gaussian mixture model and EM algorithm for approximate inference.
•
Incremental calculation of score metric for updating Bayesian network structure.
•
Evolving Bayesian network structure using hill-climbing algorithm.
•
Hill-climbing algorithms that explores multiple peaks parallel.
•
Updating score metrics based on changed edges only.

Abstract

In the Big Data era, large volumes of data are continuously and rapidly generated from sensor networks, social network, the Internet, etc. Predicting from online event stream is an important task since users usually need to predict some future states and take some actions in advance. Many applications need online prediction models which can evolve automatically with data distribution drift and algorithms which can support single-pass processing of data, which are still faced with many challenges. In this paper, the authors propose a predictive complex event processing method based on evolving Bayesian networks. The Bayesian model is designed based on event type and time with inference method based on Gaussian mixture model and EM algorithm. When learning the structure of Bayesian network from event streams, this method supports calculating score metric incrementally when new data is arrived or edges in the network are changed. Evolving Bayesian network structure is supported based on hill-climbing method. The system can continuously monitor the Bayesian network model and modify it if it is found to be not appropriate for the new incoming data. The method of this paper is evaluated in road traffic domain with both real application data and data produced by a simulated transportation system. The total percentage error is 8.12% for real data and 7.78% for simulated data, while the best result for other methods is 11.79% for real data and 14.59% for simulated data. The experimental evaluations show that this method is effective for predictive complex event processing and it outperforms other popular methods when processing traffic prediction in intelligent transportation systems.

Introduction

In the Big Data era users require new technology to process data stream with high speed and variety of data type. Besides the traditional stream processing technology, it is important to catch the relations inside online streaming data. Most of the data streams are composed of primitive events that produced by sensor networks, internet, social networks, etc. The semantic information inside primitive events is quite limited. In real application, people usually pay more attention to higher level information such as business logic and rules. For example, enormous events are generated in a trading system, but a fraud detecting system only care about the events that can cause frauds. Complex Event Processing (CEP) [1] is the technology that interprets and combines streams of primitive events to identify higher level composite events. CEP has been used in many areas, such as sensor networks for environmental monitoring, continuous analyzing of stocks to detect trends, etc.

Predictive complex event processing is the technology to identify events before they have happened, so that they can be eliminated or their effects mitigated [2]. For example, a financial institution wishes to detect frauds or a financial regulator wishes to catch illegal trading patterns in advance. Another example is to predict the traffic status in road networks and take some actions proactively to mitigate or eliminate undesired future states. Simple predictive CEP can be supported by rule-based method which means users define some patterns and the system then continuously monitor the event streams to predict future events. However, for complex cases it is not easy to define predictive CEP patterns exactly. In such circumstance, predictive analytics technology can be applied to support predictive CEP. Predictive analytics applies several statistical and data mining techniques such as clustering, classification, regression and so on. Recently, neural networks (especially deep neural networks) and Bayesian Networks (BN) are commonly used as prediction models. BN and it variations, such as Dynamic Bayesian Networks (DBN) [3], Adaptive Bayesian Networks (ABN) [4], etc., are widely used in predictive analytics because they have the following advantages: (i) they allow to express directly the fundamental qualitative relationship of direct causation, (ii) there exists mathematical methods to estimate the state of certain variables given the state of other variables, (iii) there are methods in order to explain to the user how the system came to its conclusions [5].

Currently predictive CEP with predictive analytics has some challenges. First, traditional predictive analytics methods are designed for database which means they assume all data is available at any time. However, in predictive CEP the system can only process data on single-pass and cannot control over the order of samples that arrive over time. Stream-based predictive analytics methods are needed which is more difficult. Second, the distribution of data can change over time. A model learned from historical data may not fit the new coming data well. This means real time modeling and learning algorithms are needed. In order to support predictive CEP with BN, an Evolving Bayesian Networks (EBN) model is needed that can adjust itself automatically when the distribution of data changes. There are some works on learning BN model incrementally but they simply assume all data is in memory and add new incoming data into existing data in each step. Finally, event streams often have high incoming rate, especially the event streams produced by wireless sensor networks. Predictive CEP is usually used to support online decision support system which need high performance even for large scale distributed event streams.

When used to handle the predictive CEP issue, the main weakness of the current predictive analytics methods is that they cannot get acceptable result due to the distribution drift of streaming event data. To address the challenges and overcome the weakness of current work, in this paper the authors propose a Predictive Complex Event Processing method based on Evolving Bayesian Networks (PCEP-EBN). The main contribution of this paper includes the following:

-
The authors propose a predictive CEP framework based on BN model which uses Gaussian mixture model and EM algorithm for approximate inference.
-
When learning and updating the BN structure, the authors propose incremental calculation methods for the score metric which support single-pass processing of event stream.
-
The authors propose algorithms to support evolving BN structure based on hill-climbing method. The hill-climbing algorithm of this paper supports exploring multiple peaks on parallel and updating the score metric based on changed edges only. Evolving BN parameters is also supported using an incremental EM algorithm for Gaussian mixture model.

The authors evaluated the PCEP-EBN method in road traffic domain with both real application data and data produced by a simulated transportation system. The results show that PCEP-EBN is an effective method for predictive complex event processing and it outperforms other popular methods when processing traffic prediction in intelligent transportation systems.

Section snippets

Complex event processing and predictive analytics

CEP is a technique that recognizes complex events based on a set/sequence of occurrences of single events by continuously monitoring the event stream. Etzion et al. defined the basic concepts and architecture of complex event processing in their book [2]. The main component in CEP is Event Processing Agent (EPA) which can generate a set of complex events as output from a set of input events by applying some logic. Event Processing Network (EPN) is composed of a collection of EPAs, event

System architecture and basic CEP

The system architecture is shown in Fig. 1. Primitive event streams generated by various event sources are processed by CEP engine. The BEPA means basic event processing agent which recognize complex event based on patterns from event stream. Multiple EPAs are connected by event channels (ECs) to create an EPN. The skeleton in PEPA is a statistical summary of the dataset which contains all information necessary to calculate the scores of the Bayesian network. The predictive complex events

Bayesian network structure learning

Definition 5 evolving Bayesian network, EBN

An evolving Bayesian network is a Bayesian network whose structure and parameters are changed according to the drift of data distribution. An EBN is represented by 〈G(t), Θ(t)〉 where G(t) and Θ(t) are the structure and parameters at time t respectively.

The EBN includes evolving of BN structure and parameters. For structure evolving, since search-and-score method is more suitable for evolving than constraint-based method, the authors use a search-and-score method with BDe metric [18] score

Evaluations of EBN

The authors evaluated the EBN method separately since it is the key part of this paper. The BN structure in Fig. 2 is a special structure and the authors can't find real data with corresponding result BN structures for accuracy evaluation. Therefore it is reasonable to draw samples from a known BN, apply structure learning on the synthetic data, and compare the learned structure with the original one. The original BN structure contains 783 event types (nodes) and 1752 edges. Since the authors

Conclusion and future work

In this paper the authors propose a predictive complex event processing method based on evolving Bayesian networks. The Bayesian model has two dimensions: event type and time. The Gaussian mixture model and EM algorithm are used as approximate method to infer the Bayesian model. This method supports calculating score metric incrementally. Bayesian network structure and parameters evolving algorithms are proposed. The evaluation results in road traffic domain with both real application data and

Acknowledgments

This project is supported by the “Study of Proactive Complex Event Processing for Large-scale Internet of Things” project of National Natural Science Foundation of China (61371116).

References (36)

E. Castillo et al.
Predicting traffic flow using Bayesian networks
Transp. Res. Part B
(2008)
S. Acharya et al.
Causal network construction over event streams
Inf. Sci.
(2014)
Y. Wang et al.
Complex event processing over distributed probabilistic event streams
Comput. Math. Appl.
(2013)
X. Zeng et al.
Incremental partial least squares analysis of big streaming data
Pattern Recognit.
(2014)
J. Zhao et al.
Temporal variations of reference evapotranspiration and its sensitivity to meteorological factors in Heihe River Basin, China
Water Sci. Eng.
(2015)
D.C. Luckham
The Power of events: An Introduction to Complex Event Processing in Distributed Enterprise Systems
(2002)
O. Etzion et al.
Event Processing in Action
(2010)
H.C. Cho et al.
Nonlinear networked control systems with random nature using neural approach and dynamic Bayesian networks
Int. J. Control, Autom. Syst.
(2008)
A. Pascale et al.
Adaptive Bayesian network for traffic flow prediction
J.R. Alcobe
Incremental hill-climbing search applied to Bayesian network structure learning

G. Sharon et al.

Event-processing network model and implementation

IBM Syst. J.

(2008)

E. Wu et al.

High-performance complex event processing over streams

J. Agrawal et al.

Efficient pattern matching over event streams

L.J. Fülöp et al.

Predictive complex event processing: a conceptual framework for combining complex event processing and predictive analytics

G. Tóth et al.

Complex event processing synergies with predictive analytics

W. Huang et al.

Deep architecture for traffic flow prediction - deep belief networks with multitask learning

IEEE Trans. Intell. Transp. Syst.

(2014)

Y. Lv et al.

Traffic flow prediction with big data: a deep learning approach

IEEE Trans. Intell. Transp. Syst.

(2015)

S. Zhu et al.

A Bayesian network Model for traffic flow estimation using prior link flows

(2013)

Cited by (26)

Few-shot node classification on attributed networks based on deep metric learning for Cyber–Physical–Social Services
2023, Pattern Recognition Letters
In Cyber–Physical–Social Systems (CPSS), the interactions among various entities form complex graphs. Many tasks can be formulated as instances of node classification. Node classification on graphs has attracted increasing research interest. However, the performance of existing graph neural networks for few-shot node classification has not achieved satisfactory results due to the limitation of the number of labeled instances. Therefore, we propose a graph-weighted prototype scaling network (GWPSN) based on deep metric learning for addressing the graph few-shot node classification problem. Specifically, we first extract node representations for the attributed graph via the simplifying graph convolutional network. At the same time, learning the importance of each node in the attributed graph is used to aggregate class prototypes. Finally, the class of the test node can be predicted by comparing the scaled metric distance between the test node and the class prototype. Experiments indicate that GWPSN can achieve superior performance on three real-world datasets and thus can provide enhanced few-shot classification services for CPSS.
Adaptive and personalized user behavior modeling in complex event processing platforms for remote health monitoring systems
2022, Artificial Intelligence in Medicine
Citation Excerpt :
Bayesian networks are one of the famous AI methods that can be used aligned with CEP. In [23], the authors proposed a complex predictive event processing method based on the evolution of the Bayesian networks. The Bayesian model is designed by an inference method based on the Gaussian mixed model and EM algorithm and based on event type and time.
Taking care of people who need constant care is essential and its cost is rising every day. Many intelligent remote health monitoring systems have been developed from the past till now. Intelligent systems explainability has become a necessity after the worldwide adoption of such systems, especially in the health domain to explain and justify decisions made by intelligent systems. Rule-based techniques are among the best in terms of explainability. However, there are several challenges associated with remote health monitoring systems in general and rule-based techniques, specifically. In this research, an adaptive platform based on Complex Event Processing (CEP) has been proposed for user behavior modeling to provide adaptive and personalized remote health monitoring. This system can manage a massive amount of data in real-time utilizing the CEP engine. It can also avoid human errors in setting rules thresholds by extracting thresholds from previous data using JRip rule-based classifier. Moreover, a feature selection method is proposed to decrease the high number of features while maintaining accuracy. Additionally, a rule adaption method has been proposed to cope with changes over time. Additionally, a personalized rule adaption method is proposed to address the need for responsiveness of the system to the special requirements of each user. The experimental results on both hospital and activity data sets showed that the proposed rule adaption method improves the accuracy by about 15 % compared to non-adaptive systems. Additionally, the proposed personalized rule adaption method has an accuracy improvement of about 3 % to 6 % on both mentioned datasets.
Evolving Dynamic Bayesian Networks by an Analytical Threshold for Dealing with Data Imputation in Time Series Dataset
2022, Big Data Research
Citation Excerpt :
Nowadays in the Big Data Era, data are rapidly and continuously generated forming large datasets from many heterogeneous sources [1].
Datasets with multiple variables are useful to identify trends promptly that can be used to support planning and decision making. However, it is quite common for these datasets to miss observations, in turn influencing the quality of the predictions. This is a challenging problem that is attracting increasing research. Although several imputation methods exist, most imputing methodologies are centred around a specific application, and these approaches fall short of desired goals when applied to time series datasets. In this sense, this paper describes a method for using evolving Dynamic Bayesian Networks by an analytical threshold for dealing with data imputation in time series datasets. Our method smoothly converges to a robust data imputation model and constantly adapts to the arrival of new data, obtaining more reliable network models. We conducted tests using real and simulated datasets. To insert missing values, we consider missing completely at random (MCAR) and missing not at random (MNAR). For each dataset, 10%, 20%, 30% and 40% missing rate was used. For each missing data rate, we inserted the missing values in 40% of the total intervals of each dataset. We benchmarked the results against widely used imputation methods on time series datasets. There is a significant improvement with reductions of values observed for imputation performance metrics NRMSE, MAE and MedAE compared with other existing methods.
Towards a control-as-a-service architecture for smart environments
2021, Simulation Modelling Practice and Theory
Citation Excerpt :
In a smart home scenario, this type of solution can detect the thermal discomfort of residents, either through temperature and humidity data or by the number of people occupying the place in a given time interval, and reacts by turning on an air-conditioning. A proactive CEP can identify, eliminate, or mitigate the effects of future events by applying forecasting and decision techniques using machine learning, data mining, or other approaches [22,26]. This type of application allows, for example, to detect that the consumption of electricity in a smart home will exceed the defined limits to ensure the comfort of residents.
The popularization of the Internet of Things has opened opportunities for applications in various economic sectors, allowing them to monitor and control various types of environments. However, with the increasing number of data sent by devices, applications have come to demand increased availability and Internet bandwidth, but these demands are still a bottleneck in regions farther from large urban centers and with less economic development. In this work, we examine how bandwidth limitations can affect cloud-based applications and propose a Fog Computing-based Control-as-a-Service architecture to dynamically process events in the context of Internet of Things and Smart Environments. The architecture is composed of a Rules Engine and a Complex Event Processor based on Supervisory Control techniques for Discrete Event Systems. The Rules Engine allows defining dynamic rules conditioning control from different types of input data. The Complex Event Processor with automatic synthesis and reconfiguration capability allows performing control actions to established rules. Finally, we present scenarios where this solution can be applied, such as Smart Farm, Smart Home, Smart City, and Smart Healthcare.
Rules engine and complex event processor in the context of internet of things for precision agriculture
2018, Computers and Electronics in Agriculture
Citation Excerpt :
CEP is a technology that allows processing, analyze and correlate large amounts of heterogeneous data in the form of events to detect patterns with direct implications in making decisions relevant to a particular domain (Boubeta, 2014). According Wang et al. (2018), CEP interprets and combines flows of primitive events, captured in most cases in real time, to identify composite events of higher level, known as event patterns or complex events. In addition, a CEP allows the processing of data flows from multiple data sources (Akila et al., 2016) and interprets a set of rules written in an ad hoc language for defining rules (Cugola and Margara, 2012).
The Internet of Things (IoT) applications monitor large data flows and events in real time, some raw data is captured from devices located in wireless sensor networks (WSN) and used to make control decisions about actuators. This can be a major problem when the devices grow in number as well as the data that is captured. In this paper, we propose an architecture called “RECEP” for the dynamic processing of events generated in the context of IoT and Precision Agriculture (PA); it is made up of two components: Rules Engine (RE) and Complex Event Processor (CEP). RE allows you to configure dynamic rules conditioning input data from different sources and planning control actions on actuators, alerts, and notifications for end users or applications. The CEP component fuses the input data at the rate at which they arrive, with the rules established in the RE and it performs a prescriptive analysis that consists not only in predicting or detecting patterns of events, but in making automatic decisions. RECEP was implemented in a virtual machine with a 1.9 GHz CPU and 6 GB RAM, then it was integrated into an intelligent irrigation system of an experimental banana plot located in Machala-Ecuador. A WSN simulator was also used to generate sensor data in large quantities, the CEP was evaluated with several test cases, and results show that it consumes computational resources with a growth trend, represented by a logarithmic regression model (r-squared > 0.9); that is, the more events are processed, there is a minimum consumption of resources. It was tested for fifteen days; around 25 thousand events/s were processed. Our RECEP can be implemented in low-cost infrastructure typical of small and large banana producers.
MARTIN: An End-To-end Microservice Architecture for Predictive Maintenance in Industry 4.0
2023, Proceedings - 2023 IEEE International Conference on Software Services Engineering, SSE 2023

View all citing articles on Scopus

View full text

Predictive complex event processing based on evolving Bayesian networks

Highlights

Abstract

Introduction

Section snippets

Complex event processing and predictive analytics

System architecture and basic CEP

Bayesian network structure learning

Evaluations of EBN

Conclusion and future work

Acknowledgments

Transp. Res. Part B

Inf. Sci.

Comput. Math. Appl.

Pattern Recognit.

Water Sci. Eng.

The Power of events: An Introduction to Complex Event Processing in Distributed Enterprise Systems

Event Processing in Action

Nonlinear networked control systems with random nature using neural approach and dynamic Bayesian networks

Int. J. Control, Autom. Syst.

Adaptive Bayesian network for traffic flow prediction

Incremental hill-climbing search applied to Bayesian network structure learning

Event-processing network model and implementation

IBM Syst. J.

High-performance complex event processing over streams

Efficient pattern matching over event streams

Predictive complex event processing: a conceptual framework for combining complex event processing and predictive analytics

Complex event processing synergies with predictive analytics

Deep architecture for traffic flow prediction - deep belief networks with multitask learning

IEEE Trans. Intell. Transp. Syst.

Traffic flow prediction with big data: a deep learning approach

IEEE Trans. Intell. Transp. Syst.

A Bayesian network Model for traffic flow estimation using prior link flows