Anomaly detection‐based intelligent computing in internet of things and network applications
Fa Zhu, Rajan Shankaran, Bin Zhao, Yongbin Zhao, Yining Gang, Xingchi Chen · Internet Technology Letters · 2021
With the popularization of smart and Internet of Things (IoT) devices, a tremendous amount of data is being produced by devices distributed everywhere on the globe. Meanwhile, many new smart applications are appearing which require Internet connection to operate. Artificial intelligence (AI)-driven IoT and network applications has been widely recognized as a promising solution for smart scenarios.1, 2 For IoT and emerging smart applications (eg, smart city, smart healthcare, and smart agriculture), it is often unfeasible to collect thousands of uniformly distributed data from heterogeneous distributed devices to train a reasonable model. Such systems typically only use relatively a small amount of samples to train a goal-oriented model which could be heavily affected by minor anomalies. Therefore, in the IoT and emerging network applications, it is urgent to detect anomalies from the collected samples to alleviate the side effect on model training.3, 4 Anomalies can be defined as the patterns that do not conform to expected behavior.5, 6 Actually, anomalies can appear in different forms in real applications, f.i., illegal intrusions on the Internet of services,7 irregular behaviors in the scenario of smart city,8 and abnormal events in smart agriculture.9 Anomaly detection has been researched for several decades and many anomaly detection methods have been proposed so far. However, an important open question to be answered is how to use anomaly detection methods to find minor meaningful patterns in real applications. This special issue will focus on identifying minor anomalies in real applications by using state-of-the-art anomaly detection methods. Examples of applications can be found in many fields, f.i., Internet of Things (IoT), Internet of Medical Things (IoMT), Smart Cities, Smart Grids and Energy Internet, Smart Healthcare, Smart Exercise Monitoring, Smart Online Education, Smart Agriculture, Ubiquitous Patient Monitoring, Intelligent Computing System, Decision Support and Therapy Improvement of AI-based Monitoring System, Medical Implant Security, Mobile Edge Network Security. The key issues when using intelligent computing to enhance IoT applications and services are the data collected by the associated applications and devices. The quality of the collected data are strongly affect the performance of intelligent computing. Additionally, some IoT applications pay more attention to minor anomalies, such as emergent event, unusual visiting. Thus, it needs to consider to improve the quality of collected data as well as the minor anomalies when adopting intelligent computing for IoT applications and services. It is unavoidable that the data collected from IoT applications and devices contains some noises or outliers which seriously impact on the quality of the collected data. The applications cannot achieve the expected results if directly using the intelligent computing methods which are not robust to noises or outliers. Another way is to adopt a outlier or anomaly detection method to remove the noises or outliers before training intelligent model for smart IoT applications or services. In many IoT applications or services, we only care about minor emergent events which can be regarded as the anomalies. For instance, we are interested in minor illegal access during Internet of services, minor patients in Internet of Medical Things, minor accidents in smart transportation, to name just a few. In order to solve this issues, it is worth to introduce anomaly detection methods into IoT applications and services. This special issue comes really timely for relevant research communities, and provides a valuable contribution to this emerging field. The selected the papers are based on their scientific quality and the strength of their contributions. We cannot argue that they cover all the topics of the domain. However, we are delighted to present a consistent special issue, covering a large set of problems in the Internet of Things and associated smart applications. The first paper "Smart Abnormal Emotion Analysis from English Sentences" establishes a system to identify abnormal emotion in the messages by using multivariate Gaussian model, which can provide constructive suggestions for company or government in time. The second paper "Detecting the Athlete's Abnormal Emotions before Competition via Support Vector Data Description" proposes a framework to monitor athlete's emotion to ensure that athlete can maintain a high level of competition. The third paper "Monitoring Physiological Status during Mid- and Long-distance Running for Smart Healthcare" proposes a framework to monitor the physiological status during sports. The framework is used to keep people be health during sports. The fourth paper "Smart Face Identification via Improved LBP and HOG Features" proposes a framework to identify face automatically, which can be used in smart applications as smart face identification part. The fifth paper "Risk Evaluation for C2C E-commerce via an Improved Credit Counting Method" proposes an intelligent risk evaluation method which can avoid anti-credit speculation and periodic deception. The sixth paper "Abnormal Pattern Detection in PPG Signals during Sports and Exercises" utilizes photoplethysmogram (PPG) signal to detect abnormal pattern during sports and exercises. The seventh paper "Semi-supervised Learning Induced Abnormal Emotional Tendency Analysis in British Culture" adopts semi-supervised learning to detect negative sentiment, which can provide useful advices for the company or government. The eighth paper "Action Recognition and Correction by Using EMG Signal for Health Sports" proposes a framework to recognize and correct the actions in sports by using EMG signals. First, the EMG signals are divided into fixed-size segments. Then, the EMG signal segments are processed by short-time Fourier transform as spectrograms. Lastly, the spectrograms are used to train a convolutional neural network to recognize actions during sports. The ninth paper "Spatio-temporal Topic Model Induced Semantic Function Measurement for Urban Landscape Design in Smart City" proposes semantic function measurement for smart city by using spatio-temporal topic model. Compared with previous work, the authors consider both spatial and temporal features of the documents. The tenth paper "Network Security and User Abnormal Behavior Detection by Using Deep Neural Network" proposes a framework to detect abnormal or illegal visiting in network by using deep neural network. The proposed framework implements fine-grained analysis for network traffic and user behavior management. The eleventh paper "Sport Training Action Correction by Using Convolutional Neural Network" proposes a framework to estimate the positions of bone joint points and postures by using a convolutional neural network. The positions of bone joint point prediction can be further used to correct the actions during sports. The twelfth paper "ECG Signal Analysis for Fatigue and Abnormal Event Detection during Sport and Exercise" proposes a framework to avoid the injury caused by the fatigue in excessive sports. The framework first utilize wearable device to collect ECG signal, then remove the noises in collected ECG signal, lastly learn a WOC-SVM model by using denoised signals. The thirteenth paper "Personalized Recommendation by Using Fused User Preference to Construct Smart Library" fuses user preference to recommend potential books or magazines for readers. The proposed recommendation method can be used to as a part of smart library. The fourteenth paper "Automatic Machinery Fault Detection via Using Distributed Sensor Information" proposes a framework to automatic detect the machinery fault. In the framework, first the mechanical parameters are collected by distributed sensors; then the collected information is used to train a generative adversarial networks to construct complete training set for learning; lastly, the complete training set is used to train an intelligent model for fault detection. The fifteenth paper "Sport Action Recognition by Fusing Multi-source Sensor Information" adopts body area network to collect individual information for recognizing and correcting actions during exercises and sports. This work can provide useful suggestions and advices for coaches to improve the efficiency of training. The sixteenth paper "On Tracing Abnormal Links in Cold Chain Logistics of Agricultural Products by Using Location Based Service" proposes a framework for abnormal link detection for cold chain logistics to ensure traceability of problematic foods during transport. The seventeenth paper "Smart Access Control System by Using Sparse Representation Features" utilizes sparse representation for face identification to implement non-contact authentication, which can avoid the infection of Covid-19. The eighteenth paper "Correlation Analysis for Illegal Tampering Image Evidence Detection" proposes a framework for illegal tampering image detection by collecting fingerprint from PRNU sensor. The fingerprints of images are analyzed by correlation coefficient. The nineteenth paper "On Monitoring Abnormal Physiological State Detection of Athletes from Internet of Bodies" utilizes Internet of Bodies to collect physiological signals, such electrocardiogram (ECG). The collected physiological signals are used to monitor physical status during sports for avoiding potential accidents. The twentieth paper "Intelligent Task Scheduling of Distributed Wireless Sensor Network in Monitoring Building Environment" constructs a wireless sensor network (WSN) to implement task scheduling for indoor environment of smart buildings. The task allocation strategy in WSN is designed by considering computable complexity and dynamic scheduling for the dynamic task scheduling problem. The twenty-first paper "Wearable sensors for smart abnormal heart rate detection during skiing" adopts ballitocardiogram signal to analyze abnormal heart rate. The ballitocardiogram signals are collected from wearable device during skiing. The guest editors are thankful to all reviewers for their efforts during reviewing the manuscripts. A special thank Editor-in-Chief Prof. Luigi Alfredo Grieco for his supportive guidance throughout the process.