Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 55729 (19)
- University: Sharif University of Technology
- Department: Computer Engineering
- Advisor(s): Hessabi, Shahin
- Abstract:
- Today, data analysis has become a valuable tool in various fields in order to discover the forms or abnormal behavior of the system. For example, detecting credit card fraud, network intrusion, or detecting defects in sensors. In this thesis, a method for fault analysis is presented, which can be used as a monitoring tool to monitor the data that is received in a time series from an Internet of Things system and understand the behavior of the system. This method is designed in such a way that by matching the components with the input data, in general, it is applicable for our desired time series. The presented algorithm is implemented in an unsupervised environment using a statistical approach to detect and analyze system behavior. The proposed algorithm works by fitting a statistical model to a training dataset of a certain size. The result of this fitting is the calculation of control limits for the features extracted from the data. It is said that the current behavior of the system is abnormal or that a fault has occurred in the system when a characteristic deviates from the calculated control limit. These control limits are updated as new data is received to accommodate the latest normal system behavior. The time complexity of the proposed algorithm is linear in terms of the size of training data and execution, which makes it suitable for execution in an online environment with limited resources.The algorithm is evaluated using time series data provided by Madden at the Intel Berkeley Research Laboratory, which includes laboratory values from a number of sensors.In this training data set, abnormal behaviors that include common abnormal behaviors and uncommon abnormal behaviors have been added manually, so that the algorithm can be evaluated. The presented algorithm has shown that it is able to identify abnormal behaviors with an accuracy of 94% and a low false positive rate. However, real data was not available and the abnormal behavior of the system was simulated. Therefore, it is not possible to say with certainty how the performance of the presented algorithm is in reality.The implemented method shows promising results, but there is a need to have real labeled data in order to provide a more accurate evaluation of this algorithm.
- Keywords:
- Anomaly Detection ; Monitoring System ; Internet of Things ; Error Analysis ; Fault Detection
Digital Object List-
محتواي کتاب - view
Bookmark