Titanic Survival Prediction

Machine Learning project predicting passenger survival in the Titanic disaster using demographic and travel information.

The goal is to explore the dataset, engineer meaningful features, and build predictive models capable of estimating survival probability.

---

Problem

The sinking of the RMS Titanic is one of the most famous maritime disasters. Using passenger information such as age, sex, class, and ticket data, the objective is to predict whether a passenger survived.

This project applies data analysis and machine learning techniques to identify the most important factors influencing survival.

---

Dataset

Dataset from Kaggle:

Titanic - Machine Learning from Disaster

Features include:

• Passenger class
• Name
• Sex
• Age
• Number of siblings/spouses aboard
• Number of parents/children aboard
• Ticket number
• Fare
• Cabin
• Port of embarkation

Target variable:

• Survived (0 = No, 1 = Yes)

---

Project Structure

titanic-ml-project/

data/

notebooks/
    01_eda.ipynb
    02_feature_engineering.ipynb
    03_modeling.ipynb

src/
    preprocessing.py
    features.py
    model.py
    train.py

reports/

---

Exploratory Data Analysis (EDA)

Key insights discovered during analysis:

• Women had a significantly higher survival rate than men.
• First-class passengers survived more frequently than third-class passengers.
• Children had higher survival probability.
• Passengers traveling in small families had better survival rates.
• Passengers embarking at Cherbourg showed higher survival rates, likely due to a higher proportion of first-class passengers.

Visualizations:

---

Feature Engineering

Several new features were created to improve model performance:

Family size

FamilySize = SibSp + Parch + 1

Passengers traveling in small families showed higher survival rates.

Has cabin

train["HasCabin"] = train["Cabin"].notna().astype(int)

Passengers with recorded cabin information had a higher chance of survival.

Deck extraction

train['Deck'] = train['Cabin'].apply(lambda x: x[0] if pd.notnull(x) else 'U')

Deck information was extracted from the Cabin column.

Categorical encoding

train = pd.get_dummies(train, columns=["Embarked"], drop_first=True)

Categorical variables were transformed using one-hot encoding.

Fare per person

train['Fare_Per_Person']=train['Fare']/train['FamilySize']

This feature provides a more accurate representation of individual socioeconomic standing by normalizing costs across family units.

Is alone

train['IsAlone'] = (train['FamilySize'] == 1).astype(int)

While FamilySize provides granular data, this feature simplifies the feature space, allowing the model to focus on the significant survival gap between these two primary categories without the noise of specific family counts.

---

Models Tested

The following models were evaluated:

• Random Forest
• Gradient Boosting
• AdaBoost
• XGBoost
• Voting Classifier

Model performance was evaluated using accuracy on validation data.

---

Results

Best performing model:

Voting Classifier with AdaBoost and XGBoost

Cross-validation accuracy: 82% - 84%
Kaggle Leaderboard: 78%

Most important features:

• Sex
• Pclass
• Fare
• Age
• FamilySize
• HasCabin

---

How to Run the Project

Clone the repository:

git clone https://github.com/rodrigofl-dev/titanic-ml-project.git

Install dependencies:

pip install -r requirements.txt

Run the training pipeline:

python src/run.py

This will:

1. Load raw data
2. Perform preprocessing
3. Create features
4. Train the model
5. Generate predictions

Predictions will be saved in:

data/submissions/submission.csv

---

Technologies Used

• Python
• pandas
• numpy
• scikit-learn
• seaborn
• matplotlib
• Jupyter Notebook

---

Author

Rodrigo Lopes - Data Science project developed as part of a machine learning portfolio.