How a small R&D team achieved great results in the Kaggle competition without using ML algorithms
A few months ago, Navigine R&D team started participating in Indoor Location & Navigation competition from XYZ10 and Microsoft Research.
The purpose of the competition was as follows: organizers provided the participants with logs from hundreds of shopping centers, where visitors walked along specific routes. It was necessary to predict the floor and the x, y coordinate of the visitor on that floor based on radio signals and inertial sensors. In addition to the logs, the participants had metadata in the form of a map, which shows the location of cafes, shops, and other points of interest (POI).
What steps have been taken by our team? Let’s take a closer look at this from their view.
- The first step was converting the data to our format. This step was necessary because the format provided by Kaggle did not quite fit us, and this applied to both logs and floor plans.
- Next, we found out that our submissions were tested by organizers only on a small subset of malls, so we were able to discard about ~70% of the data and only looked at those few for the rest of the competition.
- Following, we looked at the floor plans and the trajectories that people walked. Logs were collected through…
