Mathematical Principles and Practices for Internet of Things Data Analysis using Machine Learning Approach
Loading...
Date
2019-10-20
Authors
Sulaiman, Batoul
Journal Title
Journal ISSN
Volume Title
Publisher
جامعة النجاح الوطنية
Abstract
Internet of Things (IoT) environment generates the data
continuously, these data need to be collected, analyzed in order to trigger
an action. For many IoT applications, the locations information for the
collected data is considered very important information, so a lot of
localization techniques in wireless sensor networks (WSNs) are used for
obtaining such information. In this work, we propose an artificial Feed Forward Neural Network (FFNN) for the IoT sensor node localization. The
proposed method was performed in many heterogeneous WSNs in which
the anchor nodes distributed uniformly. Matlab software was used to
implement this network which has a single hidden layer with 20 neurons.
Two different training algorithms were used to evaluate this network which
are the Levenberg-Marquardt algorithm and Gradient descent algorithm.
The estimated locations for the sensor nodes obtained from the proposed
FFNN (single layer) was compared with the results obtained from another
structure for the network which is the Deep feed-forward (multilayer)
neural network. Also, a comparison with a known localization algorithm
'Weighted centroid algorithm based on RSSI' was performed. Our results
showed that the feedforward (single layer) neural network is a good
IX
localization approach which gives us the most accurate estimated locations
for the sensor node in WSNs