Anomaly Detection and Complex Event Processing Over IoT Data Streams: With Application to eHealth and Patient Data Monitoring

Contents
List of figures
List of tables
Preface
Biographies
1 IoT data streams: concepts and models
1.1 IoT streams in the context of Big Data
1.2 Static vs. continuous data systems
1.3 Time variability in data streams
1.4 Dynamic data stream structure – the drift concept
1.4.1 Drift speed
1.4.2 Drift severity
1.4.3 Drift influence zones
1.4.4 Drift occurrence frequency and recurrence
1.4.5 Drift predictability
1.4.6 Drift concept in real-world application
1.5 IoT data streams in healthcare
1.5.1 Healthcare concepts
1.5.2 IoT in healthcare applications
1.6 Key features
References
2 Data stream processing: models and methods
2.1 Semantic primitives for stream processing
2.1.1 Nonblocking operations
2.1.2 Blocking operations
2.2 Window-based methods
2.3 Feature domain processing
2.3.1 Basic magnitude features and time-locked averaging
2.3.2 Template matching
2.3.3 Weighted moving averages: frequency filtering
2.3.4 Frequency-domain
2.3.5 Time-frequency domain
2.4 Dimensionality reduction and analysis techniques
2.4.1 Dimensionality reduction
2.4.2 Indexing
2.4.3 Sequence matching
2.4.4 Clustering
2.4.5 Classification
2.5 Key features
References
3 Anomaly detection
3.1 Introduction to anomaly detection
3.2 Challenges in anomaly detection
3.3 Anomaly types and detection techniques
3.3.1 Anomaly types
3.3.2 Anomaly detection mode
3.3.3 Anomaly detection techniques
3.3.4 Anomaly detection in health domain
3.4 Accuracy and prediction from anomaly detection
3.4.1 Classification and regression metrics
3.4.2 Anomaly detection rules on data streams
3.4.3 Other metrics
3.5 Key features
References
4 Complex event processing
4.1 Fundamental concept of CEP
4.2 Primitive functions for CEP
4.3 MultiIoT data stream in healthcare
4.3.1 Chronically ill patient monitoring
4.3.2 Dementia care, IoT data stream and CEP
4.3.3 Other application scenarios
4.4 CEP and AI
4.4.1 CEP video processing
4.4.2 CEP video and audio processing in healthcare applications
4.5 Key features
References
5 Rule-based decision support systems for eHealth
5.1 Introductory concepts and background in expert systems and decision support systems
5.2 Clinical Decision Support (CDS) basics
5.2.1 Knowledge-based DSS (KB-DSS)
5.2.2 Nonknowledge based DSSs
5.2.3 Hybrid -intelligent- DSS
5.2.4 Examples of medication-related CDSS
5.3 Implementation challenges of a DSS
5.3.1 Other challenges
5.3.2 Outlook on the precision medicine initiative
5.4 Rule base systems in practice and their limitations
5.4.1 Semantic data enrichment and rule base systems
5.4.2 Limitations of rule base system for IoT stream processing
5.4.3 Limits of processing at scale
5.4.4 Limits of processing unseen information from the stream
5.4.5 Limits of processing multimodal data streams of continuous variables
5.5 Key features
References
6 Integrating technological solutions into innovative eHealth applications
6.1 Telemedicine network architectures
6.2 Telemedicine healthcare services
6.2.1 Asynchronous healthcare services
6.2.2 Synchronous healthcare services
6.3 Telemedicine healthcare application
6.3.1 Physiological parameter monitoring
6.3.2 Chronic disease monitoring
6.3.3 Disease detection
6.3.4 Disease diagnostics
6.3.5 Disease prediction
6.3.6 Disease treatment
6.3.7 Multicondition application
6.4 Ambient Assisted Living (AAL)
6.5 Telemedicine rehabilitation
6.6 Patient-physician interactions
6.7 Complex event processing for remote patient monitoring
6.8 Other related issues to patient monitoring
6.8.1 NoSQL patient-electronic-record
6.8.2 Privacy and security
6.9 Key features
References
7 IoT, edge, cloud architecture and communication protocols
7.1 IoT architecture
7.2 Decentralized architecture paradigms
7.3 IoT architecture components
7.4 IoT stream processing infrastructures
7.5 Criteria for IoT architecture selection
7.6 Application protocols
7.6.1 Constrained Application Protocol (CoAP)
7.6.2 Message Queue Telemetry Transport (MQTT)
7.6.3 Extensible Messaging and Presence Protocol (XMPP)
7.6.4 Advanced Message Queuing Protocol (AMQP)
7.6.5 Data Distribution Service (DDS)
7.7 Infrastructure protocols
7.7.1 Routing Protocol for Low Power and Lossy Networks (RPL)
7.7.2 6LowPAN
7.7.3 IEEE 802.15.4
7.7.4 Bluetooth low energy
7.7.5 EPCglobal
7.7.6 LTE-A (Long Term Evolution – Advanced)
7.7.7 Z-Wave
7.8 Comparison of infrastructure protocols
7.9 Criteria for communication protocols selection
7.10 Key features
References
8 Machine learning
8.1 Learning models
8.1.1 Supervised learning
8.1.2 Unsupervised learning
8.1.3 Reinforcement learning
8.2 Hybrid learning models
8.2.1 Semisupervised learning
8.2.2 Self-supervised learning
8.2.3 Multiinstance learning
8.3 Statistical inference
8.3.1 Inductive learning
8.3.2 Deductive learning
8.3.3 Transductive learning
8.4 Learning techniques
8.4.1 Multitask learning
8.4.2 Active learning
8.4.3 Online learning
8.4.4 Transfer learning
8.4.5 Ensemble learning
8.5 Federated learning
8.5.1 Basic concepts
8.5.2 Federated efforts in data-driven medicine
8.5.3 Current FL efforts for digital health
8.5.4 Technical considerations
8.6 Handling concept drifts
8.7 ML frameworks
8.7.1 ML frameworks without special hardware support
8.7.2 Interactive data analytic and visualization tools
8.7.3 Other data analytic frameworks and libraries
8.7.4 Deep learning frameworks and libraries
8.7.5 Machine learning and deep learning frameworks and libraries with MapReduce
8.8 Key features
References
9 Anomaly detection, classification and CEP with ML methods
9.1 Anomaly detection by deep learning methods
9.1.1 Deep learning based feature extraction
9.1.2 Working principles of learning methods
9.1.3 Learning feature representations of normality
9.1.4 Anomaly measure-dependent feature learning
9.1.5 End-to-end anomaly score learning
9.1.6 End-to-end machine learning pipeline for medicine
9.1.7 Deep complex anomaly detection
9.2 Complex event processing
9.2.1 Three-level data fusion model
9.3 Key features
References
10 Architectures and technologies for stream processing
10.1 Introduction case study
10.2 Ingestion and communication system - Kafka
10.2.1 Kafka architecture
10.2.2 Kafka-concepts
10.3 Communication protocol between producer devices and the Kafka ingestion system - MQTT
10.4 Stream processing and single-stream event detection - Faust
10.5 Complex event processing - Kafka Streams with KSQL
10.5.1 Kafka stream concept
10.5.2 Kafka stream architecture
10.5.3 KSQL
10.6 Other processing platform
10.6.1 Batch processing
10.6.2 Stream processing
10.7 Frameworks used in healthcare
10.8 Key features
References
11 Technical design: data processing pipeline in eHealth
11.1 Medical background of ECG data
11.2 ECG data sets
11.3 Dataset used in the case study
11.3.1 Class imbalance
11.4 Pipeline: preprocessing module
11.4.1 Denoising approaches review
11.5 Pipeline: core-processing module
11.6 Pipeline: anomaly detection, classification and complex event processing
11.6.1 Anomaly detection
11.7 Pipeline: classification and prediction
11.7.1 Challenges ECG classification
11.7.2 Types of ECG classifications
11.7.3 Selected model for the case study
11.8 Key features
References
12 Working procedure and analysis for an ECG dataset
12.1 Processing and analysis of an ECG dataset by Faust cluster computing
12.1.1 Trade-off 1: HTM anomaly threshold vs. data size vs. detection rate
12.1.2 HTM anomaly detection and segmentation Python script
12.1.3 Faust signal extraction and classification evaluation
12.1.4 Processing time along various layers of the architecture
12.1.5 Classification
12.1.6 Faust worker Python script for signal extraction and classification
12.1.7 Directory tree explanation and execution instructions
12.2 Event processing network diagram
12.3 Data representation and enrichment
12.3.1 Linked data and ontologies for healthcare applications
12.3.2 Proposed semantic representation along the data pipeline
12.3.3 Issues in the context of semantic enrichment
12.4 Key features
References
13 Ethics, emerging research trends, issues and challenges
13.1 Ethics and privacy in patient data monitoring
13.1.1 Ethical considerations for machine learning healthcare applications
13.1.2 EU General Data Protection Regulation (GDPR)
13.1.3 Machine learning for diagnosis and treatment in the context of data privacy
13.2 Noninvasive and personalized solutions for elderly based on IoT technologies
13.3 Detection vs prediction eHealth solutions at scale
13.3.1 Preventive, predictive, personalized, and participatory (P4) medicine
13.3.2 Predictive, personalized, preventive and participatory (P4) medicine applied to telemedicine and eHealth
13.3.3 Towards interpretable modeling approaches
13.3.4 Hybrid machine learning and mechanistic models
13.3.5 Controlling critical transitions in patient trajectories
13.4 Key features
References
Index