The following repository provides my personal approach to the Spaceship Titanic Kaggle Challenge.
My best submission was created using the GradientBoostingClassifier and a lot of experimenting and reached over 80.8% (Top 6%, place 157/2455 on 09.08.22). The GB-Classifier included in the notebook is reproducible and scores over 80.4% on the test data.
Note: Only two of the models explored during the modeling process are covered in the notebook.
Example plots from the statistical analysis
As part of the analysis, I will not only take a look at the provided features alone, but will also discover relationships amongst variables, engineer new features and investigate different strategies to fill missing values in the dataset. I will describe very detailed what I am analyzing, which insights the different plots provide, why I choose certain values for missing data imputation, etc.
The goal of the whole process is to gain a deep understanding of the different features and of course to clean the dataset, before starting to engineer models to predict the target variable "Transported".
The modeling process itself is not the main goal of this notebook. But in the end of it, in section 5, I will also cover two example baseline ensemble models, one RandomForest and one GradientBoostingClassifier, both scoring over 80% on the test data. The hyperparameter values used were defined by tuning the models via GridSearchCV, which is not part of the jupyter notebook.