A Body Sensor Network Data Repository With A Different Mining Technique
A Body Sensor Network Data Repository With A Different Mining Technique
A Body Sensor Network Data Repository With A Different Mining Technique
ISSN: 22503676
Volume - 2, Issue - 1, 105 109
Assist. Professor, Dept. of ECE. Universal College of Engg. Tech, Guntur Dt., A.P, India, pratyuchaitanya@gmail.com 2 PG scholar, Dept. of ECM. K.L.University, Vaddeswaram, Guntur Dt., A.P, India, prathyushakuncha@gmail.com
Abstract
To get efficient use of large amounts of body sensor data, a human movement data using clustering, and a technique to analyze sensed physiological signals are used. Recent years have witnessed a large influx of applications in the field of cyberphysical systems. An important class of these systems is body sensor networks (BSNs) where lightweight embedded processors and communication systems are tightly coupled with the human body. One major obstacle is managing repositories that store the large amount of sensing data. To address this issue, we propose a data mining approach inspired by the experience in the areas of text and natural language processing. BSNs can provide researchers, care providers and clinicians access to tremendously valuable information extracted from data that are collected in users natural environment. With this information, one can monitor the progression of a disease, identify its early onset, or simply assess users wellness. We represent sensor readings with a sequence of characters, called motion transcripts. Transcripts reduce complexity of the data significantly while maintaining morphological and structural properties of the physiological signals. These transitions are efficiently captured using the concept of n-grams. To facilitate a lightweight and fast mining approach, we reduce the overwhelmingly large number of n-grams via information gain (IG) feature selection. To further take advantage of the physiological signals structure, our data mining technique focuses on the characteristic transitions in the signals
Index Terms: Body sensor networks (BSNs); data mining; n-grams, BSN Data Repository.
---------------------------------------------------------------------***---------------------------------------------------------------------1. INTRODUCTION
Body sensor networks (BSNs) are becoming an increasingly popular field of research for a variety of applications ranging from fall and posture detection [2], [5] and telemedicine to rehabilitation and sports training [3], [4]. These systems are composed of lightweight wearable sensors that capture different physiological data from the human body. This physiological data may include an inertial description of human movements, electrocardiograph (ECG) readings of the human heart, electromyography (EMG) readings of the muscle activity, skin conductance level, blood pressure, and many more. Physiological signals of these modalities can be observed by sensors mounted on wearable devices. However, the modern sensing platforms are not perfect. Along with the useful physiological information they also capture noise and other data collection artifacts. Data collection artifacts are abnormalities in the signal that can be introduced by the specific sensor deployment conditions. For example, the type of a strap used to attach a sensor to the body can significantly affect the recorded observation. Additionally, from the high level perspective, similar movements may look the same, and the specific movement execution can introduce variations in the sensors observations of those movements. For example, a sit-to-lie movement can be performed smoothly, or the subject can throw themselves on the bed and briefly bounce on the mattress. While both movements achieve the same goal, from the inertial sensor perspective they do not look exactly the same.
M. CHAITANYA SUMAN* et al. [IJESAT] INTERNATIONAL JOURNAL OF ENGINEERING SCIENCE & ADVANCED TECHNOLOGY example, blinking), inability to finish some movements, impaired balance while walking, muscle rigidity, and varies tremors, severely affect human movements and can be observed with the help of inertial sensors. The task is aided by the fact that many devices in our daily lives, such as cell phones, already have inertial sensors built-in. Furthermore, the seamless nature of BSN nodes allows their deployment prior to serious health problems to monitor the onset of the condition. The following example demonstrates the usefulness of the idea. A person can be monitored for an extended period of time (e.g., multiple years) with the help of a few sensor nodes. All of the data are simply collected and stored in the data repository. After some time, this person is diagnosed with a disease that involved gait abnormalities. It would be beneficial to analyze old data and extract gait parameters for disease evaluation, examination of disease progress, and treatment. A data mining approach would be able to identify movements of interest, in this case walking, so that the raw data of movements can be used to extract the required gait parameters.
ISSN: 22503676
Volume - 2, Issue - 1, 105 109
because it suggests that while observations may not match in their entirety, due to data collection artifact and individual subject performance, they still have a significant structural similarity and can be compared by extracting representative signal properties. For structural techniques to be effective it is essential to design an effective data representation approach that simplifies the multidimensional BSN data yet captures the structure of the signal.
4. DESIGN CHALLENGES
BSN sensor nodes are highly constrained in terms of memory, processing resources, and battery lifetime. This means that all of the collected data cannot be stored on the wearable device, communicated wirelessly for an indefinite amount of time, or processed with complicated and possibly slow computational approaches on the device itself. At the same time, they have a potential to produce very large data sets over time. This suggests that the data representation approach needs to significantly reduce the complexity of the data, while maintaining the characteristic structure of the signal. This task is further complicated by the possibility of errors in the signal and intersubject variability in movement performance. This problem can be solved by applying limited processing to the sensor data, as it is being collected, to reduce its size and complexity. This step, however, needs to preserve the structural parameters of the signal. This can be achieved by applying limited processing that exclusively focuses on identifying transitions in the signal that uniquely characterize each movement. For this step to be successful, it is essential for the system to extract the properties of the signal capable of capturing such characteristic transitions. While in other systems, redundancy may be acceptable and even desirable, the resource and time constraints of the BSNs demand that the considered set of signal properties be minimal. That means that some of the machine learning and signal processing techniques may not be suitable for implementation on the sensor nodes. With these requirements in mind, we present a data mining model for large BSN data repositories. Our approach has two steps: 1) We first propose a solution for the problem of data representation. For this, we define a technique for movement primitive construction from multidimensional physiological signals using clustering algorithms. We explore algorithms that preserve the original structure of the signal, even if human movements have timing inconsistencies. Unlike other works, we consider multiple clustering techniques for primitive construction using a small and computationally simple feature set. We combined the constructed primitives with their timing
M. CHAITANYA SUMAN* et al. [IJESAT] INTERNATIONAL JOURNAL OF ENGINEERING SCIENCE & ADVANCED TECHNOLOGY properties to generate string transcripts to capture the relational information from the signal. 2) We then define a novel data mining model that explores structural and relational properties of the string transcripts. We use information gain (IG) to select the parts of transcripts that can best differentiate between movements, and then define a tree-based classifier for data indexing and mining. We verify the quality of our model by applying it to a pilot movement data set. While other works focus on achieving the highest possible accuracy of classification, the key objective of our work is to define a data mining approach that can be applied to a very large data set, which results in decisions that favor speed and simplicity of computation. While we do not explicitly trade the accuracy for speed and simplicity, it is an important tradeoff we keep in mind during system design decisions.
ISSN: 22503676
Volume - 2, Issue - 1, 105 109
5.2 Hardware
During the experiment, subjects were equipped with several TelosB sensor nodes with custom-designed sensor boards. Each sensor board has a tri-axial accelerometer (providing x-, y-, and z-axis of acceleration) and a bi-axial gyroscope (providing x- and y-axis of angular velocity). Sensors were sampled at 50 Hz. This sampling frequency is high enough to provide acceptable resolution of the movements, and has been previously suggested by several other authors for physical movement monitoring applications [9], [6]. Furthermore, it satisfies the Nyquist criterion [11]. After collecting the data, each node sent its readings to the base station. In our case, the base station is a node without a sensor board, which forwards all of the received data to the PC via a USB connection for further processing.
5. PROBLEM OVERVIEW
In this section, we first describe the system used throughout this work. We then discuss the desired properties of the solution in the context of the defined system. Finally, we briefly introduce the pilot application used to evaluate our approach.
M. CHAITANYA SUMAN* et al. [IJESAT] INTERNATIONAL JOURNAL OF ENGINEERING SCIENCE & ADVANCED TECHNOLOGY
ISSN: 22503676
Volume - 2, Issue - 1, 105 109
7. CONCLUSION
We generated motion primitives based on instantaneous simple features and unsupervised clustering. We showed how the signal primitives can be combined into motion transcripts, which are unidimensional representations of the multidimensional BSN data. Inspired by the techniques of natural language processing, we applied the concept of ngrams retrieval for tracking transitions in the movement transcripts. Due to the large number of n-grams extracted from a movement trial, we apply a simple IG approach to the features to select k features that provide the most information about each sensing axis. Based on the selected n-grams we build a suffix tree for fast query and identification of movements in the database. We demonstrate that the system can achieve average F-score of 97% on our pilot data with the help of only one characteristic for each movement transition. We also explored the tradeoff between the length of the extracted n-grams and the required number of features for the best classification results.
REFERENCES
[1] H. Ghasemzadeh, V. Loseu, and R. Jafari, Collaborative signal processing for action recognition in body sensor networks: A distributed classification algorithm using motion transcripts,[ in Proc. 9th ACM/IEEE Int. Conf. Inf. Process. Sensor Netw. 2010, pp. 244255. Fig. 2. TelosB sensor node with a custom sensor board.
M. CHAITANYA SUMAN* et al. [IJESAT] INTERNATIONAL JOURNAL OF ENGINEERING SCIENCE & ADVANCED TECHNOLOGY [2] B. Lo, J. Wang, and G. Yang, BFrom imaging networks to behavior profiling: Ubiquitous sensing for managed homecare of the elderly,[ in Proc. 3rd Int. Conf. Pervasive Comput., May 2005, pp. 101104. [3] D. Brunelli, E. Farella, L. Rocchi, M. Dozza, L. Chiari, and L. Benini, BBio-feedback system for rehabilitation based on a wireless body area network, in Proc. 4th Annu. IEEE Int. Conf. Pervasive Comput. Commun. Workshops, Mar. 2006, pp. 527531. [4] M. Lapinski, E. Berkson, T. Gill, M. Reinold, and J. Paradiso, BA distributed wearable, wireless sensor system for evaluating professional baseball pitchers and batters,[ in Proc. Int. Symp. Wearable Comput, 2009, pp. 131138. [5] A. Volmer, N. Kruger, and R. Orglmeister, BPosture and motion detection using acceleration data for context aware sensing in personal healthcare systems, in Proc. World Congr. Med. Phys. Biomed. Eng., Munich, Germany, 2009, pp. 71 74. [6] R. Mayagoitia, A. Nene, and P. Veltink, BAccelerometer and rate gyroscope measurement of kinematics: An inexpensive alternative to optical motion analysis systems. Biomech., vol. 35, no. 4, pp. 537542, 2002. [7] D. Gelb, E. Oliver, and S. Gilman, BDiagnostic criteria for Parkinson disease, Archives Neurol., vol. 56, no. 1, pp. 3339, 1999. [8] V. Shnayder, B. Chen, K. Lorincz, T. Fulford-Jones, and M. Welsh, BSensor networks for medical care,[ in Proc. 3rd Int. Conf. Embedded Netw. Sensor Syst., 2005, pp. 314327. [9] N. Ravi, N. Dandekar, P. Mysore, and M. Littman, BActivity recognition from accelerometer data,[ in Proc. Nat. Conf. Arti. Intell, 2005, vol. 20, no. 3, pp. 15411546. [10] H. Bunke and A. Sanfeliu, Syntactic and Structural Pattern Recognition: Theory and Applications. Singapore: World Scientific, 1990. [11] N. Stergiou, Innovative Analyses of Human Movement. Champaign, IL: Human Kinetics, 2004.
ISSN: 22503676
Volume - 2, Issue - 1, 105 109