Here the link to my notebook in which I did preprocessing on a dirty Titanic dataset retrieved from Kaggle and then there's a brief proof that making predictions on a good dataset increase the performance of the model with low efforts.
https://colab.research.google.com/drive/1TmJWPPk6E3AKkLBNzcsvgF95hdGOx3BA?usp=sharing
Dataset attached below.