Data Fault Detection Using Multi-label Classification in Sensor Network
Detection of data fault is a hit point in sensor network in recent years and multiple data faults may occur at the same time. However the existing detection methods can only detect one type of data fault at the same time. To solve this problem, the detect
- PDF / 366,577 Bytes
- 11 Pages / 439.37 x 666.142 pts Page_size
- 51 Downloads / 387 Views
Abstract Detection of data fault is a hit point in sensor network in recent years and multiple data faults may occur at the same time. However the existing detection methods can only detect one type of data fault at the same time. To solve this problem, the detection of data fault is model as multi-label classification task, where an instance may belong to more than one label. We firstly break down the sensor data into instances using sliding window and then employ feature extraction methods to extract data features from instances. After this we associate each instance with corresponding data fault labels. A comparative experimental evaluation of five multi-label classification algorithms is conducted on aforementioned multi-label dataset using a variety of evaluation measures. Experimental results and their analysis show preliminarily that: (1) LP outperforms other classification algorithms on all evaluation measures since it takes the correlation of multiple data fault labels into account in detection of data fault; (2) outlier is the easiest data fault to detect and high noise is the hardest data fault to detect.
Keywords Algorithms Classification Multi-label classification
Fault detection
Sensor network
Z. Zhang (&) S. Li Z. Li School of Computer Science, Northwestern Polytechnical University, Xi’an, China e-mail: [email protected] S. Li e-mail: [email protected] Z. Li e-mail: [email protected]
Z. Wen and T. Li (eds.), Practical Applications of Intelligent Systems, Advances in Intelligent Systems and Computing 279, DOI: 10.1007/978-3-642-54927-4_10, Springer-Verlag Berlin Heidelberg 2014
101
102
Z. Zhang et al.
1 Introduction Sensor network, a data-centric network, has received worldwide attention in recent years and has a lot of important applications such as military [1], agriculture [2], architecture [3], and so on. Sensor network is used to sense and collect information from physical world, which provides people with significant evidences when trying to make meaningful conclusions. Therefore, the data sensor network collected must be true and reliable, which is the key that whether sensor network has been applied successfully or not in practical deployments. If there are some faulty sensor readings in collected data, it is possible to make an improper decision or draw meaningless conclusions. Several practical deployments have observed faulty sensor readings caused by improper data calibration, or hardware failures, or low battery levels, or interference around environment, and so on [4, 5]. To ensure the data quality of the collected data, fault detection in sensor network have been an important research direction [6]. Noise is an important factor of affecting data quality. Eiman [7] proposed a Bayesian approach, combining prior knowledge of true sensor readings, the noise characteristics of the sensor and the observed noisy readings, to reduce the uncertainty due to noise and enhance the reliability of sensor readings. Spatialtemporal correlation is an important character
Data Loading...