WiFi Localization

Kaggle Competition

In this project you will use a widely used dataset in a Kaggle competition to implement a localization app in an SDN controller that oversees a Wi-Fi Network. The use case is very real and such applications are very critical for commercial and non-commercial purposes. For example Splunk is extensively used as a monitoring system that can aggregate client RSSI measurements from Wi-Fi APs around the campus. Such aggregations are used by educational institutions at the moment to supplement COVID19 contact tracing.

Data

The paper describing the dataset can be obtained from here

Deliverables

What you will need to deliver:

1. Replicate a baseline localization algorithm such as shown in the competition notebooks or found in Github (e.g. in https://github.com/sharan-naribole/wlan_localization). You can pick a localization method that you understand.
2. Document how the algorithm in (1) works.
3. Convert the dataset to a tracking dataset (i.e. use the timestamps) that emulates users moving along feasible paths/trajectories. A feasible path is a path that each consecutive location is a neighbor.
4. Attempt to localize the trajectory a hypothetical user goes through assuming knowledge of the indoor map. You are free to use explicit probabilistic reasoning approaches such as Hidden Markov Models (HMMs) or, if you are familiar with, black box approaches such as Convolutional or Recursive Neural Networks (eg. https://arxiv.org/pdf/1810.07377.pdf)
5. Submit your code in a Github repo and be prepared to be called to answer questions on your methods selected. Any student unable to answer conceptual questions about the localization method(s) involved will be automatically assigned a grade of 0.