What it does

Given a dataset (coded under assumption of .csv files) of device signals with cell towers and individual app usage per device, predicts future data usage/metrics for a given cell tower - this feature can be compounded across many cell towers in an area, and over time, an effective heat map can be produced of the activity per cell tower in order to more exactly determine what cell towers need more servicing depending on the time of day.

How we built it

We use Machine Learning algorithms (Kalman filter) to predict the data consumption on the mobile network and potentially advise carriers where to add temporary antennas to boost the network or to overlap cell towers to better provide service to customers. PCA is also optionally applied to project multiple variables, not just data usage, into one variable as a better metric instead.

On the given data set (the code of which is currently tested on one network, one day) we sum all the data usage over all devices that connected to a cell tower every hour, and plot this - we use the Kalman filter to predict the behavior of the data usage, which can be useful for spotting odd occurrences/events (e.g. natural disasters) but also as a better metric for providing better services in the eyes of the carrier.

Challenges we ran into

Parsing the given dataset with pandas (instead of w/ SQL, or even Mongo if the data weren't relationally defined) was the only problem, as we stumbled along the way in joining tables and grouping by intervals, etc.

When producing the MVP for the project, we went back and forth on deciding whether to use hidden Markov models or neural networks for the predictive algorithm - we ended up going with HMMs but scrapping it midway through the hackathon for Kalman filters, which we believe to be much more expressive.

What's next for SmartCells

The originally intended stage two would be to produce a network optimization algorithm to see what cell towers can be overlapped (by increasing signal) or added in order to better service customers in the eyes of the carrier.

Share this project: