Wi-Fi Navigation Using Machine Learning With Consideration Of Cyclic Dynamic Behaviour
Wi-Fi based localization using machine learning has been proven to be an attractive approach for finding the location prediction with avoidance of accumulation of errors as other sensors such as odometry and inertial sensing. Researchers have developed various models to predict locations based on tr...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English |
| Published: |
2020
|
| Subjects: | |
| Online Access: | http://eprints.utem.edu.my/id/eprint/25388/1/Wi-Fi%20Navigation%20Using%20Machine%20Learning%20With%20Consideration%20Of%20Cyclic%20Dynamic%20Behaviour.pdf http://eprints.utem.edu.my/id/eprint/25388/2/Wi-Fi%20Navigation%20Using%20Machine%20Learning%20With%20Consideration%20Of%20Cyclic%20Dynamic%20Behaviour.pdf http://eprints.utem.edu.my/id/eprint/25388/ https://plh.utem.edu.my/cgi-bin/koha/opac-detail.pl?biblionumber=119762 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Wi-Fi based localization using machine learning has been proven to be an attractive approach for finding the location prediction with avoidance of accumulation of errors as other sensors such as odometry and inertial sensing. Researchers have developed various models to predict locations based on trained machine learning. A site survey is typically performed to collect fingerprints and a neural network is trained on those fingerprints. The trained model is then placed into operation. However, dynamic changes in the location and navigation behavior of users make the prediction process more challenging in terms of accurate prediction of location. One common mobility behavior of navigation runs is the cyclic dynamics or re-visiting the same place more than one time. Most machine learning models, developed for location prediction, lack sufficient handling of dynamic changes or leveraging them for better predictions. To fill this gap, this study builds a new simulator with two components: one for incorporating dynamic information of navigation in given Wi-Fi dataset and using them to generate the corresponding time series of any navigation run, it is named as Wi-Fi Simulator for Cyclic Dynamic (Wi-Fi-SCD) while the other is useful for converting any dataset to time series with cyclic dynamics, it is named as Cyclic Dynamic Generator (CDG). Furthermore, in this study, two novel location prediction machine learning models were developed. The first is Knowledge Preservation Online Sequential Extreme Learning Machine (KP-OSELM) and the second is Infinite Term Memory-based Online Sequential Extreme Learning Machine (ITM-OSELM). The KP-OSELM model is distinctive from other models cited in the literature, because it preserves knowledge gained in certain areas to restore again when the person re-visits the area again. In KP-OSELM, knowledge is preserved within the neural network structure and is enabled based on feature encoding. The ITM-OSELM model is distinctive from other models cited in the literature, because it carries external memory and transfers learning to preserve old knowledge and restoration. ITM-OSELM is more efficient than KP-OSELM when the percentage of active features is low. Meanwhile, KP-OSELM does not require any external blocks to be added to the neural network (unlike ITM-OSELM), which makes it much simpler. In area based scenarios, KP-OSELM and ITM-OSELM both achieved accuracies of 68%. Moreover, when evaluating KP-OSELM and ITM-OSELM on Wi-Fi-SCD, for three navigation cycles, the highest accuracies achieved were 92.74% and 92.76%, respectively. However, the execution time of KP-OSELM was 1176 second while much less time was needed for ITM-OSELM to be executed with a value of 649 second. Furthermore, when evaluating KP-OSELM and ITM-OSELM on CDG, for three cycles, 100% accuracy was achieved for both models. As a conclusion, this study has provided the literature of machine learning in general and WiFi navigation in particular with various models to support the localization without any restriction on the type of Wi-Fi that is used and with consideration of the practical and dynamic behaviors that can be leveraged to improve the localization performance. |
|---|