Titanic Dataset EDA & Insights

Why the Titanic Dataset Remains a Gold Standard for EDA

The Kaggle Titanic dataset is often a developer's first foray into structured data science. While seemingly compact, it provides exceptional real-world complexity: missing age distributions, categorical embarkation points, high-cardinality ticket labels, and nested family relationships.

In this project, my objective was to extract actionable survival patterns and establish clean data preprocessing pipelines.

Handling Missing Values & Feature Engineering

Approximately 20% of the Age column and over 75% of the Cabin column were missing. Rather than dropping rows indiscriminately, systematic imputation was implemented:

Title Extraction: Extracted honorific titles (Mr., Mrs., Miss., Master., Dr.) from passenger names.

Median Age Imputation: Imputed missing ages using the median age of the passenger's specific title group.

Family Size Feature: Combined SibSp (siblings/spouses) and Parch (parents/children) to calculate overall family size.

python

import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt

df = pd.read_csv('titanic.csv')

# Extract title from passenger name
df['Title'] = df['Name'].str.extract(' ([A-Za-z]+)\.', expand=False)
title_age_medians = df.groupby('Title')['Age'].median()

# Impute age based on title
df['Age'] = df.apply(
    lambda row: title_age_medians[row['Title']] if pd.isnull(row['Age']) else row['Age'],
    axis=1
)

# Engineer FamilySize
df['FamilySize'] = df['SibSp'] + df['Parch'] + 1
df['IsAlone'] = (df['FamilySize'] == 1).astype(int)

Key Visual Discoveries

Using Seaborn boxplots, heatmaps, and faceted bar plots, key demographic drivers emerged:

Gender Disparity: Female passengers achieved an overall survival rate of approximately 74%, compared to just 19% for male passengers.

Class Stratification: Passengers in 1st class survived at a rate of 63%, whereas 3rd class passengers experienced only 24% survival.

Solo Travelers: Solo travelers in 3rd class suffered the highest mortality rate.

This exercise proved that deep exploratory data analysis and domain-aware feature engineering are more influential in predictive workflows than algorithm selection alone.