Team_AI1_8

Titanic Survival Prediction using Logistic Regression

👥 Team Members

• Sanket Savagar - Data handling and Preprocessing
• Prajwal - Feature Engineering and encoding
• Ranjit - Model training and Evalution
• Anurag - Reedme file

---

🎯 Problem Statement

The objective of this project is to predict whether a passenger survived the Titanic disaster or not using machine learning techniques. This is a binary classification problem where the output is either:

• 1 → Survived
• 0 → Not Survived

---

📊 Dataset Description

We used the Titanic dataset from Kaggle, which contains information about passengers such as:

• Age
• Gender
• Passenger class
• Fare
• Family details

The dataset is divided into:

• “train.csv” → used for training the model
• “test.csv” → used for testing

---

🧹 Data Preprocessing

Before training the model, we cleaned and prepared the data:

• Removed unnecessary columns:

  • Name
  • Ticket
  • Cabin
  • PassengerId

• Handled missing values:

  • Age → filled using median
  • Embarked → filled using mode
  • Fare → filled using median

---

⚙️ Feature Engineering

We created new features to improve model performance:

• FamilySize = SibSp + Parch + 1
• IsAlone:
  • 1 → passenger is alone
  • 0 → passenger has family

---

🔢 Data Encoding

Categorical variables like gender and embarked location were converted into numerical values using:

• One-hot encoding (pd.get_dummies)

This is required because machine learning models only understand numerical data.

---

✂️ Train-Test Split

The dataset was divided into:

• 80% Training data
• 20% Testing data

This helps evaluate how well the model performs on unseen data.

---

📏 Feature Scaling

We applied StandardScaler to normalize the data so that all features have similar scale, which improves model performance.

---

🤖 Model Used

We used Logistic Regression, which is suitable for classification problems.

Reasons:

• Works well for binary classification
• Simple and efficient
• Provides good accuracy

---

📊 Model Training & Validation

• Model trained using training dataset
• Applied Cross Validation (cv=5) to check consistency
• Compared training and testing accuracy

---

📈 Model Evaluation

We evaluated the model using:

• Accuracy Score → measures correctness
• Confusion Matrix → shows correct & incorrect predictions
• Classification Report:
  • Precision
  • Recall
  • F1-score

---

📉 Results

• Training Accuracy ≈ 80–85%
• Testing Accuracy ≈ 75–80%
• Cross Validation Accuracy ≈ 78–80%

👉 This indicates the model performs well and generalizes properly.

---

🔗 GitHub Collaboration

• Repository created on GitHub
• All team members were added as collaborators
• Each member contributed to different parts of the project
• Changes were managed using commits and pull requests

---

🧠 Conclusion

This project demonstrates how machine learning can be used to solve real-world classification problems. By applying preprocessing, feature engineering, and Logistic Regression, we successfully built a model that predicts Titanic passenger survival with good accuracy.

---

🚀 Future Improvements

• Use advanced models like Random Forest or Decision Trees
• Hyperparameter tuning
• Add a web interface using Flask

---