I've worked on a similar approach for Indoor Locationing, I had 5 APs with OpenWRT reading the RSSI of my phone and some Neural Nets implementation, which eventually learned where I am in the house.
If something simple like the location of a AP changes, would this system have to train from scratch with the entire dataset?
NN can be trained online.
this is really interesting, company I work for has recently started working on "location analytics" using signal strength (rssi) of the signal sent when your phone scans for wifi