We were provided telemetry data from CUAUV, Cornell's underwater rover team. One interesting objective was, given some definition of "anomalous readings", to determine and predict when the robot would encounter such errors.
What it does
As the data itself is "unlabeled", our model takes several assumptions about the data and does its best to come up with generalizations. The model assumes and attempts to handle four types of "anomalies": a permutation of
How we built it
A significant portion was completed with Microsoft Azure ML Studio - it was intuitive to use especially for trying out novel ideas and models. Also, we used python notebooks to visualize certain parts of the data.