Créer l'environnement virtuel avec Python 3

python3 -m venv kaggle-titanic

Activer l'environnement

source kaggle-titanic/bin/activate

Installer les dépendances

Sauvegarder les dépendances dans requirements.txt

pip freeze > requirements.txt

Lancement de jupyter

jupyter notebook --port=8891

Ignorer les dossiers/fichiers a pas push (.gitignore)

Titanic Survival Prediction - Random Forest Classifier

Table of Contents

• Project Description
• Dataset
• Preprocessing
• Feature Engineering
• Model
• Model Evaluation
• Final Submission
• Requirements
• How to Run

Project Description

This project aims to predict the survival of passengers aboard the Titanic using machine learning techniques. Using a Random Forest Classifier, we predict whether a passenger survived or not based on various features like age, sex, class, and family size. The model is trained and evaluated using cross-validation, and the final model is used to predict survival for the test dataset.

Kaggle Score:

As of now, the best score achieved on Kaggle is 0.79665.

Dataset

The dataset used for this project is the Titanic dataset, which is available on Kaggle. It consists of two files:

• train.csv - Contains the training data with both features and the target variable (Survived).
• test.csv - Contains the test data with features only (no target variable), used for generating predictions.

Preprocessing

The preprocessing steps are as follows:

1. Reading the Data: We load the training and test datasets from CSV files using pandas.
2. Handling Missing Values:
   • We use SimpleImputer to fill missing values in the Age column with the median of that column.
   • We handle other missing values by filling them with 0.
3. Categorical Variables:
   • We use pd.get_dummies() to convert categorical variables like Embarked, Sex, and Cabin into numeric features.
4. Feature Engineering:
   • We create a new feature called FamilySize by combining the SibSp (siblings/spouses aboard) and Parch (parents/children aboard) columns.
   • We create an IsAlone feature to identify whether the passenger is traveling alone (FamilySize == 1).
   • We create a HasCabin binary feature to indicate whether a passenger has cabin information available.

Feature Engineering

• Family Size: The size of the passenger's family is derived by summing SibSp and Parch and adding 1 (for the passenger).
• IsAlone: This is a binary feature indicating whether a passenger is traveling alone. If FamilySize equals 1, then the passenger is alone.
• HasCabin: A binary feature to indicate whether the passenger's cabin information is missing or not.

Model

We use the Random Forest Classifier, a powerful ensemble learning method based on decision trees:

• Hyperparameters:
  • n_estimators = 100: The number of trees in the forest.
  • max_depth = 10: The maximum depth of the trees.
  • random_state = 3: Ensures reproducibility.

The model is evaluated using 5-fold cross-validation to estimate its generalization performance.

Model Evaluation

We use cross_val_score to perform 5-fold cross-validation and obtain an average accuracy score. This method divides the data into 5 subsets (folds), trains the model on 4 folds, and tests it on the remaining fold, repeating the process for each fold.

The mean accuracy from cross-validation is printed to the console, allowing us to evaluate the model's performance.

Mean cross-validation score:  0.8170736300295023
Kaggle website score :  0.79665