Titanic Dataset - Data Analysis

Project Purpose

Predict whether passengers survived or not using the Titanic dataset. Steps such as data preprocessing, feature engineering, hyperparameter tuning and model optimization were taken to increase the success of the model.

Dataset Used

The dataset contains passenger information from the Titanic disaster. Each passenger has a survival status (Survived) and various characteristics (age, gender, ticket class etc.).

Dataset Columns:

• PassengerId: Passenger ID
• Survived: 0 = dead, 1 = alive
• Pclass: Ticket class (1 = 1st class, 2 = 2nd class, 3 = 3rd class)
• Name: Passenger name
• Sex: Passenger gender
• Age: Passenger age
• SibSp: Passenger number of siblings/spouses
• Parch: Passenger number of parents/children
• Ticket: Ticket number
• Fare: Ticket fare
• Cabin: Passenger cabin
• Embarked: Passenger port of embarkation (C = Cherbourg, Q = Queenstown, S = Southampton)

What was done

1. Data Exploration and Preprocessing

Missing Data Detection and Filling:

• Missing data were filled using methods such as mean, mode and median.

New Features Added:

• FamilySize: Family size (SibSp + Parch).
• IsAlone: 1 if the passenger is alone, 0 otherwise.

Numericalization of Categorical Values:

• Categorical variables such as Sex and Embarked were converted to numerical values ​​using the Label Encoding method.

2. Feature Selection and Model Training

• Important features were identified using Recursive Feature Elimination (RFE). The best 5 features were selected: Pclass, SibSp, Parch, FamilySize, IsAlone.
• Logistic Regression model was created and trained on training data.

3. Model Optimization

• Hyperparameter Tuning: The hyperparameters of the model were optimized using GridSearchCV and RandomizedSearchCV.

• Data Scaling: Data was standardized using StandardScaler to make the model work faster and more efficiently.

4. Model Performance Evaluation

• The success of the model was evaluated with metrics such as Accuracy, Precision, Recall, F1-Score.

• The classification success of the model was visualized using Confusion Matrix.

Libraries Used

• Pandas: Data manipulation and analysis
• Numpy: Numerical computations
• Scikit-learn: Model training, hyperparameter tuning, data preprocessing
• Matplotlib and Seaborn: Data visualization