Skip to content

My solution for the Kaggle competition Porto Seguro Safe Driver Prediction. This solution uses an Ensemble of different algorithms and ranked 138 out of 5169 (Top 3%).

Notifications You must be signed in to change notification settings

pklauke/Kaggle-PortoSeguroSafeDriverPrediction

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Kaggle-PortoSeguroSafeDriverPrediction

This repository contains my solution for the Porto Seguro Safe Driver Prediction competition: https://www.kaggle.com/c/porto-seguro-safe-driver-prediction. This solution ranked 138 of 5170.

The goal of this competition was to find out whether a driver will make a claim or not. Due to the fact that the data was anonymized feature engineering was difficult. Therefore my solution focuses mainly on emsembling a variety of models using weighted blending.

The majority of models were trained in the notebook Predictor . This includes the gradient Boosted Decision Trees LightGBM, XGBoost, CatBoost and the Regularized Greedy Forest rgf_python. For more variety 3 modified versions of public kernels from other Kaggle users were used aswell. These kernels used LightGBM, XGBoost, rgf_python and keras.

In addition the Field-Aware Factorization Machine LibFFM was trained in the notebook LibFFM Predictor . The used library is a modified LibFFM version that supports early stopping with patience.

All models were trained using mostly 5-fold cross validation multiple times. Each model was trained each time on different cross validation random seeds. The predictions of all runs of a model were averaged. The averaged predictions were blended in the last step.

The models were ensembled in the notebook Ensembling . The ensembling method used was weighted blending. A self-written optimization algorithm was used to optimize the weights using the out-of-fold predictions. This algorithm decided not to use XGBoost and CatBoost for the final submission.

About

My solution for the Kaggle competition Porto Seguro Safe Driver Prediction. This solution uses an Ensemble of different algorithms and ranked 138 out of 5169 (Top 3%).

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published