ICACT20240199 Slide.18
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Thank you very much for your time, I am willing to answer any questions that you might have
|
|
ICACT20240199 Slide.17
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Reference
|
|
ICACT20240199 Slide.16
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
In conclusion, this paper compares SVM and Single-layered Perceptron for sensor fault detection in WSN using labeled datasets from various scenarios. High Classification Accuracy: Both SVM and Single-layered Perceptron achieved impressive accuracy exceeding 99% across different scenarios, indicating their effectiveness in detecting sensor faults in WSN. Simulation results indicate that SVM demonstrates higher accuracy than Single-layered Perceptron, suggesting its potential superiority in sensor fault detection within WSN.
|
|
ICACT20240199 Slide.15
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Table 2 presents the performance comparison between the two machine learning methods used in sensor fault detection. In this study, testing the accuracy of the two machine learning methods is considered as the performance metric. The SVM classifier and the neural network, both two machine learning methods have achieved good classification accuracy, as shown in the figure the accuracy of all four scenarios is above 99%. Moreover, SVM has higher accuracy compared to the neural networks for the three scenarios.
|
|
ICACT20240199 Slide.14
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Table 1 presents the Single-layered perceptron testing accuracy for sensor data of the four different scenarios.
|
|
ICACT20240199 Slide.13
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Fig. 8 presents the SVM classification of the Outdoor single-hop scenario. Fig. 9 presents the SVM classification of the Outdoor multi-hop scenario. In the figures, the decision boundary of the SVM classification has separated the normal sensor data and the faulty sensor data. Normal sensor data is shown in red color and the faulty sensor data is shown in green color.
|
|
ICACT20240199 Slide.12
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Fig. 6 presents the SVM classification of the Indoor Sensor Single scenario while Fig. 7 presents the SVM classification of the Indoor Sensor Multi-hop scenario.
|
|
ICACT20240199 Slide.11
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Data from single-hop and multi-hop wireless sensor networks are included in the dataset, which is labeled wireless sensor networks data. Specifically, the dataset from four distinct sensors has been utilized to train the machine learning models to evaluate the performance of the sensor fault detections. The four different scenarios are the Indoor Single-hop scenario, Outdoor Single-hop scenario, Indoor multi-hop scenario, and Outdoor multi-hop scenario. In a single-hop scenario, there is direct communication between the sensor and the base station, sensors can directly send their sensed data to the base station. In a multi-hop scenario, the sensed data is routed through an intermediate node between the source and the destination.
|
|
ICACT20240199 Slide.10
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Dataset and ML Approach
|
|
ICACT20240199 Slide.09
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
A publicly accessible dataset of sensor measurements was trained using the single-layered NN and SVM machine learning methods with linear kernel function. With the labels of the dataset being faulty and non-faulty consisting of two features which are the value of Temperature and the value of Humidity, for the sake of this study. These values are collected within six hours at an interval of 5 seconds.
|
|
ICACT20240199 Slide.08
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Datasets
|
|
ICACT20240199 Slide.07
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Artificial Neural Networks, including the single-layered perceptron, are well-suited for tasks involving noisy and complex sensor data. ANN learning is robust to errors in the training data, making it an ideal choice for scenarios where data quality might be compromised. The ability to learn from noisy data is particularly beneficial in applications such as sensor fault detection in Wireless Sensor Networks (WSN). The single-layered perceptron in an Artificial Neural Network utilizes the backpropagation algorithm for learning. Backpropagation employs gradient descent, a technique to fine-tune network parameters based on the sum squared training errors.
The goal of the gradient descent is to minimize the cost function by iteratively updating the weights in the direction that reduces the error between predicted and target values. This iterative process continues until the algorithm converges to a set of weights that optimally fit the training set.
|
|
ICACT20240199 Slide.06
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
SVM is particularly suited for binary classification, dealing with scenarios where there are only two classes. It relies on support vectors, which are subsets of training data close to the decision boundary. SVM uses the concept of a margin, defined as the distance between support vectors of the two classes separated by the decision boundary. The objective is to choose a decision boundary that maximizes the margin between the two classes. The goal of SVM is to achieve maximum margin classification, ensuring a larger separation between the classes.
|
|
ICACT20240199 Slide.05
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
The method used is a comparative analysis between machine learning-based algorithms for sensor fault detection in WSN: These algorithms are Support Vector Machines (SVMs) and Artificial Neural- Network (ANNs)
|
|
ICACT20240199 Slide.04
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
There are various types and causes of faults in WSNs. Sensor faults can be categorized broadly into two primary groups: faults based on location and those based on time. Data-centric faults and system-centric faults are two general categories of faults that may be distinguished based on where they occur. Sometimes faults occur temporarily and it only takes a limited duration of time. Moreover, the time-based faults are further divided into transient faults and persistent faults.
|
|
ICACT20240199 Slide.03
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
The literature on sensor fault detection in WSNs displays a varied set of strategies, including distributed, centralized, and hybrid approaches, with a notable emphasis on machine learning-based methods. This diversity highlights the complex nature of the problem and underscores the importance of flexible solutions. The review trend indicates a growing reliance on machine learning techniques for sensor fault detection. Methods such as SVMs, recurrent neural networks, K-Nearest Neighbors, and artificial neural networks have been explored. These approaches leverage collaborative sensor efforts, fault classification, and anomaly prediction to enhance the accuracy of fault detection in WSNs.
|
|
ICACT20240199 Slide.02
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
The Wireless Sensor Network system comprises several wireless devices with limited processing power, battery, and memory space used in sensing changes or occurrences in the environment. These changes could be temperature, humidity, motion, air density, wind velocity, etc. WSNs have found extensive application in diverse sectors such as the health care, military, transportation sectors and also in smart homes. As a working system, there are challenges associated with it which includes it having faulty sensors which can lead to increased network congestion and high power consumption
|
|
ICACT20240199 Slide.01
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
This presentation is sectioned into different parts namely; Introduction, Sensor Fault, Method, ML Approach, Dataset and Evaluations and Conclusion
|
|
ICACT20240199 Slide.00
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Hi everybody, my name is Israel Ehile. Our team members are mainly from the School of Engineering and Digital Sciences from Nazarbayev University, namely Dr. Refik Caglar Kizilirmak, Abubakar Abdulkarim and Israel Ehile from the Department of Electrical and Computer Engineering. I am very glad to have this opportunity to share the work of our team. This presentation is titled Utilizing Machine Learning for Sensor Fault Detection in Wireless Sensor Networks
|
