Animal_Gender

ABC-Animal-Behaviour

Predict gender of birds from their trajectories.

The code here provides all the the things which I tried.

Feature Engineering(what worked and didn't worked)

• Harvesine distance between maximum {latitude,longitude} and minimum {latitude,longitude} and multiplied it by 2( Assuming that distance will be 2 times for arrival and departure from a reference point)( Tried computing harvesine distance between each of the consequtive points and summing them up but it didn't worked due to too much noise)

• Minimum/Maximum/Mean {latitude,longitude} . (Tried variance too but it didn't worked in improving the results).

• Tried computing area of polygon of trajectory but it didn't worked so wan't included.

• Tried counting dives and elevations as a feature but didn't worked.

• Feature that worked most was length of stay(which means no of times bird stayed in journey)

• Also at the end of competition features like left to right extent (difference) gave some good improvement.

• One important thing to note here was that some of features were normalized with maximum no. of days travelled by the bird which was major part of the improvement.

Library Requirements:

Do run if not installed : pip install library name

• numpy
• pandas
• mlbox
• pyproj
• shapely
• functools
• scipy
• csv

Usage:

• Put all the data(train and test folders) inside data folder.
• To run python automl.py