Dynamic Exploratory Data Analysis with Streamlit

Upload. Explore. Visualize. Understand. All in One Click.

Abstract

The Dynamic Exploratory Data Analysis app empowers users—regardless of coding skill—to perform meaningful EDA by simply uploading a CSV file. Built with Streamlit, the app intelligently detects data types and offers insightful plots across several tabs: Univariate, Bivariate, Multivariate, and PCA.

The app allows analysts, students, and researchers to visualize trends, detect patterns, identify class imbalance, and uncover dimensionality relationships—all in a few clicks.

🧠 Background & Problem Statement

Exploratory Data Analysis (EDA) is often a repetitive and time-consuming task, especially when using notebooks. Many professionals struggle to scale it across projects or explain the results to non-technical audiences. This project eliminates those hurdles by delivering a no-code EDA dashboard that is:

• Intuitive to use
• Customizable for many data types
• Deployable on Streamlit Cloud
• Capable of both statistical and graphical insight generation

🗂 Dataset

• Primary Dataset Used: Heart Disease / Heart Failure Dataset
• Compatible Formats: Any CSV dataset with at least one categorical column
• Preprocessing Handled Automatically:
  • Detection of dtypes (numeric vs categorical)
  • Null-value inspection
  • Category count and encoding (when needed for PCA)
  • Dynamic sample adjustment for performance

⚙️ How the App Works

Once a dataset is uploaded, the app parses data and enables different analysis modules:

🏠 Home Tab

• Displays dataset structure
• Sample, filter, or remove duplicates
• Shows data types, nulls, and value counts
• Dynamic question prompts for user guidance

📊 Univariate Analysis

• Bar plots and pie charts for categorical distributions
• Boxplots and violin plots for numeric variables
• Questions prompt users to consider outliers, distribution shapes, and class balance

📈 Bivariate Analysis

• Compare two categorical variables or a categorical and numeric feature
• Automatically displays:
  • Grouped bar plots
  • Boxplots for continuous vs category
  • Count distributions with insights

📉 Multivariate Analysis

• Explore up to 3 categorical and multiple numerical variables
• Generate scatter plots, correlation matrices, and grouped bar charts

🧬 Principal Component Analysis (PCA)

• Perform PCA interactively and color projections by any categorical variable
• Displays 3D PCA plots and explained variance graphs
• Select features, choose hue and target groupings

🧪 Advanced Analysis

• Built-in KDE plots for comparing continuous variables across 2 categorical groups
• Example: Resting ECG vs Exercise Angina vs average

📈 Key Results

• ✅ Built and deployed a fully functional, public EDA tool
• 🔍 Supports datasets with any number of categorical/numeric features
• 📊 Includes rich visualizations: violin plots, scatter matrices, PCA plots, and histograms
• 📁 Users can explore data without coding, empowering broader adoption of data science

💡 Lessons Learned / Innovations

• Efficient EDA requires automated handling of mixed data types
• Realized the need to limit data size for cloud hosting (1GB RAM limit on Streamlit Cloud)
• Added question-driven guidance to help less experienced users interpret plots
• Modularized code via src/components/data_transformation.py and utils.py

💬 Discussion

This app reimagines EDA as a guided storytelling process, not just a series of plots. It works especially well for:

• Student assignments
• Hackathon data reviews
• Medical/survey data exploration
• Exploratory steps before feature engineering

With PCA and multivariate plots, users can uncover relationships even without domain knowledge.

📣 Call to Action

If you want to explore datasets visually and quickly—try the app:
👉 Launch Dashboard
Fork the GitHub repo to customize or contribute.