AXA-Driver-Telematics-Analysis

This project aim to design a solution to the AXA Driver Telematics challenge on Kaggle.

The repository includes code of my solution with additional code implementing ideas which where release after the competition had ended.

Code overview:

• scripts/sample_the_dataset.m - create positive and negative data sets for local evaluation purposes. It lets you define and import:
  • The driver of interests trips batch (positive data set)
  • The number of arbitrary irrelevant trips (negative data set)
• scripts/local_evaluation.m - use k-fold cv to assess model AUC performance
• functions/getTelematicMeasurements - this function takes the Cartesian coordinates $(X,Y)$ of a trip, rotate them with regard to the centroid mean, and calculate measurements with respect to the time dimension, such as: speed, acceleration, distance, and orientation at each sec
• functions/getSpatialMeasurements - this function takes the Cartesian coordinates $(X,Y)$ of a trip, simplified it by removing the time dimension utilizing RDP (Ramer–Douglas–Peucker), and making coordinates pairs of equally spaced length.
• scripts/preprocess_the_dataset split the 547,200 csv files into N mat files on disk for further analysis.
• unittests/test_all.m - run all unit tests

How to Run a demo

• Download the project files
• Download all the data files from Kaggle
• Run scripts/sample_the_dataset.R
• Run the different scripts and reports, especially script/local_evaluation.R

How to Run a full submission

• Download the project files
• Download all the data files from Kaggle
• Run scripts/preprocess_the_dataset.R
• Run the different models within models directory
• The submission file will appear by default at submission folder.