In this project, we present a solution to the Titanic Competition that achieves a 0.78 score on the test dataset (TOP 13% on 07/09/2023). Moreover, we present a simple but beautiful Streamlit App that allows the users to pretend they are aboard the Titanic and find out their chances of surviving the sinking. https://www.kaggle.com/code/marcpaulo/titanic-playground-for-new-kagglers-0-78
This project may be exciting for those who are starting their Data Science journey, exploring the Kaggle environment, or want to learn how to use Streamlit in their projects. There are also some TO-DO's and other ideas for you to implement in order to further your knowledge and achieve a better score in the competition.
The topics that we cover here are:
1. Exploratory Data Analysis (EDA): --> Exhaustive feature analysis, outlier detection, missing values, visualize the distribution and structure of the data, initial guess on the importance of each feature, etc. #pandas #seaborn #numpy
2. Data Preprocessing: --> Create a new feature from the existing ones (feature extraction), deal with missing data, design and implement clean, efficient, and reusable Pipelines and Data Transformers using the #sklearn library.
3. Let's train some Models: --> Test basic classification algorithms (LogisticRegression, SVC, RandomForest, GradientBoosting, KNN), and run some GridSearch with Cross-Validation for hyperparameter optimization. Plot the GridSearch results to compare the performance of all the different settings and choose the best configuration.
/data/train.csv: data used to train the models. It contains information about 891 passengers.
/images/RMS_titanic.jpg: images used to decorate the Streamlit app.
/main_app/streamlit_app.py: launches the app using the powerful Streamlit library.
/models/trained_grad_boost.pkl: trained model resulting from the training notebooks and used in the App.
/notebooks/training_playground.ipynb: long notebook implementing the Data Science part in detail.
/notebooks/training_grad_boost.ipynb: shot version of the playground notebook in which we only train and save the best model.