🚢 Titanic Survival Prediction — Logistic Regression

This project builds an interpretable Logistic Regression model to predict which passengers survived the Titanic disaster.
Using demographic, socio-economic, and travel-related features, the model uncovers survival determinants while providing strong predictive performance.

---

📌 Project Objective

The primary goal is to predict passenger survival (Survived = 1 or 0) using features such as:

• Sex
• Age
• Passenger Class (Pclass)
• Fare
• Siblings/Spouses aboard (SibSp)
• Parents/Children aboard (Parch)
• Embarkation Port
• Family size & socio-economic indicators

Logistic Regression is chosen for its transparency, interpretability, and suitability for binary classification.

---

📁 Dataset Details

• Source: Titanic Dataset (Kaggle)
• Samples: 891 rows
• Features: 11 usable predictors after cleaning
• Target: Survived (0 = No, 1 = Yes)

Key Predictors

• Sex
• Pclass
• Age
• Fare
• SibSp
• Parch
• Embarked

Preprocessing Performed

• Missing values handled (Age median imputation, Embarked mode fill)
• Dropped non-informative fields: PassengerId, Name, Ticket, Cabin
• One-hot encoding for categorical variables
• StandardScaler on Age and Fare
• Train-test split (80/20)

---

🔍 Exploratory Data Analysis — Summary

Class Distribution

• ~38% survived → moderate imbalance

Major Patterns

• Sex: Females had significantly higher survival probability
• Pclass: First-class > Second-class > Third-class
• Fare: Higher fare → higher survival odds
• Age: Younger passengers survived at higher rates
• Family Features: Small family groups slightly improved survival chances

Core Insights

• Social and economic factors strongly influenced outcomes
• Wealthier and female passengers had clear advantages
• Fare correlates positively with survival while Pclass correlates negatively

---

🧠 Model Summary

Algorithm Used:

• Logistic Regression with scaled features
• Solver: lbfgs
• Regularization: default
• One-hot encoded categorical fields

---

🎯 Model Performance

Metric	Score
Accuracy	0.80
Precision (Survived)	0.76
Recall (Survived)	0.71
F1-score (Survived)	0.74
ROC-AUC	0.86
PR-AUC	0.85

Interpretation

• Strong ability to distinguish survivors from non-survivors
• Good balance between precision & recall
• ROC-AUC 0.86 indicates reliable generalization

---

🧪 Learning & Validation Curves

• Training and validation accuracy converge around 0.80
• Low variance, minimal overfitting
• Model generalizes well without excessive complexity

---

⭐ Feature Importance (Coefficients & Analysis)

Most influential predictors:

1. Sex (female) → strongest positive effect on survival
2. Fare → higher fare passengers survived more
3. Pclass → lower class passengers had lower odds
4. Age → younger passengers slightly more likely to survive

Embarked and family-related features showed moderate or subtle influence.

---

💡 Key Insights

• Survival followed stark socio-economic and gender patterns
• Females and first-class passengers had the highest survival chances
• Higher fare values reflected priority access to rescue
• Simple linear models can still capture real-world behavioral patterns effectively

---

📦 Summary Table

Item	Value
Dataset	Titanic Passenger Dataset
Target	Survived
Best Model	Logistic Regression
Accuracy	0.80
ROC-AUC	0.86
PR-AUC	0.85
Dominant Features	Sex, Fare, Pclass, Age
Data Size	891 records

---

🧱 Tech Stack

• Python
• Pandas, NumPy
• Matplotlib, Seaborn
• Scikit-learn
• Statsmodels
• Jupyter / Google Colab