EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
3609. Enhancing Wi-Fi Fingerprinting through the Use of an Automated Supervised Autoencoder
Invited abstract in session TD-25: Applications of Machine Learning in Optimization, stream Combinatorial Optimization.
Tuesday, 14:30-16:00Room: 011 (building: 208)
Authors (first author is the speaker)
| 1. | Tiny Du Toit
|
| School of Computer Science and Information Systems, North-West University |
Abstract
Indoor localisation is an essential area with vast implications for navigation, emergency services, and a range of Internet of Things contexts. The challenge lies in the intricate nature of indoor spaces and the limitations inherent in current technologies for precise indoor localisation. This research addresses these issues using a supervised autoencoder (SAE) approach on the UjiIndoorLoc dataset, encompassing Wi-Fi fingerprint data from various buildings and levels at Jaume I University. An AutoML search was employed to identify the best hyperparameters, including learning rate, number of epochs, layers in both the autoencoder and multilayer perceptron, nodes per layer, and dropout rate. Bayesian optimisation, an efficient search strategy, was used to test different hyperparameter combinations and neural network layer configurations to find the most effective models. The findings demonstrate the model's high accuracy rates: 99.91% for building identification, 91.45% for floor identification, and an overall accuracy of 91.18%. The SAE model's performance showcases its capability and competitive edge over a standard model and various methods, underscoring its potential to overcome the difficulties of complex indoor localisation tasks.
Keywords
- Machine Learning
- Expert Systems and Neural Networks
- Telecommunications
Status: accepted
Back to the list of papers