Kaggle Titanic Competition

This notebook is about prediting the mortality ie survival of passengers of the Titanic for Kaggle. It contains a quick but still complete data science life cycle from data cleaning, feature engineering to model selection and prediction. It comprises a train-validation-test split set-up and is based on pipes to prevent data leakage. The score kaggle score so far is 78,5 percent.

Data Cleaning

Here, only three variables contain missings. Cleaning is based on standard median inputing (for continuous) and most frequent (categorical). However, a regression based approach to imputation for age is explored too.

Feature engineering

Particularly, it tackles

• the names and titles of passengers and
• the number of relatives. It starts the exploration of cabin and fare.

Model Selection

First, logistic regression is explored and its hyperparameters are tweaked. Then, the same is done for Random Forest and finally XGboost.