Grab AI SEA: Safety Challenge

This repository is the submission for the AI challenge for S.E.A hosted by Grab. The selected challenge is Safety. The challenge can be found via this website.

Author

by: Satsawat Natakarnkitkul (Net)

Email: n.satsawat@gmail.com

Country: Thailand

Motivation: This challenge is very interesting in so many ways, but as I use Grab nearly everyday. Hence this challenge is the most impact to the Grab users.

Repository structure

Notebook

• The notebook Grab AI Challenge_Safety_Data Exploration.ipynb is mainly used as part of data understanding and EDA for sensor data provided by Grab. You may not run this notebook, but it will provide some understandings and explanation onto telemetry data of the sensor world.
• The notebook Grab AI Challenge_ML model comparison.ipynb is purposely created to train and test ML techniques to produce the final model as well as try on feature engineering and other data transformation.
• The notebook Grab AI Challenge_GridSearchCV and Feature Engineering.ipynb shows the grid search for XGBoost algorithm; with the current model and feature engineering, it achieved the AUC score of 0.5945.

Model

• This folder contains the final model object to be used for prediction.

Code

• This folder contains the final python source code for manipulating, creating new features and predicting the data set.

Img

• This folder contains the image embedded onto EDA and other notebooks.

Output

• This folder contains the bookingID, predicted class, and probability of the prediction, this is the outcome from running the py script.

Run Instruction

The model is used to predict the safety of the trip as such the assumption is that this is not the real time prediction (online), but rather an offline (data for each booking ID is available). The transformation is the aggregation of each booking ID onto single observations and feed into the model for prediction.

To run the prediction, please use Safety_Prediction.py in the code directory.

1. The script in code folder will read in the feature data file within data/safety/features folder.
   • If there's any change in the data path, please adjust the DATA_DIR onto the correct folder respectively.
2. The script will automatically run the feature transformation and engineering.
3. The script will load the XGBoost model object from model directory to make a prediction.
4. The script will save the prediction with bookingID onto ../output/all_prediction.csv file.
5. If LABEL_IND = True in the script, it will attempt to run evaluation between the prediction with true label.
   • The true label file should be in the data/safety/labels folder with the proper bookingID and label columns.
   • If there's any change to labels folder, please adjust this to LABEL_DIR in the script respectively.
   • If the evaluation is not neeeded, you can turn this off by setting LABEL_IND = False in the script.