Global COVID-19 Mortality & Outcome Prediction (Python)

End-to-end data analytics and machine learning pipeline analyzing global COVID-19 trends, mortality risk, and patient outcomes

---

📌 Project Overview

This project performs a comprehensive data analytics and machine learning analysis on global COVID-19 data to understand the spread, severity, and outcomes of the pandemic across countries and regions.

Using Python, the project integrates:

• Exploratory Data Analysis (EDA)
• ETL (Extract–Transform–Load) pipelines
• Statistical analysis
• Supervised machine learning models
• Geographic and time-series visualizations

The analysis is designed to reflect real-world public-health and data-science workflows, emphasizing interpretability, scalability, and analytical rigor.

---

🌍 Business & Public Health Context

During global health crises, policymakers and healthcare systems rely on data to:

• Monitor disease spread
• Predict mortality risk
• Allocate healthcare resources
• Assess recovery trends
• Compare regional impacts

This project demonstrates how data analytics and machine learning can support evidence-based decision-making during large-scale public health emergencies.

---

🎯 Project Objectives

Analytical Goals

• Analyze global COVID-19 trends over time
• Identify relationships between confirmed cases, deaths, recoveries, and active cases
• Engineer meaningful health metrics (fatality & recovery rates)
• Predict COVID-19 mortality using machine learning
• Classify outbreak outcomes (recovery-dominant vs death-dominant)

Technical Goals

• Build a clean, reusable ETL pipeline
• Apply regression and classification models
• Evaluate model performance using multiple metrics
• Visualize global and temporal patterns effectively

---

📊 Dataset Summary

• ~49,000+ records across multiple countries
• Daily global COVID-19 reporting (2020)
• Publicly available open data sources

Key Features

• Province / State
• Country / Region
• WHO Region
• Latitude & Longitude
• Date
• Confirmed Cases
• Deaths
• Recovered
• Active Cases

Derived metrics:

• Case Fatality Rate
• Recovery Rate

---

🔄 Data Engineering (ETL Pipeline)

Extract

• Loaded raw CSV data into Pandas DataFrames

Transform

• Converted date fields for time-series analysis
• Handled missing geographic values
• Removed duplicate records
• Filtered invalid zero-case records
• Engineered health indicators:
  • Case Fatality Rate
  • Recovery Rate

Load

• Saved transformed datasets for reproducible downstream analysis

This ETL workflow ensures data quality, consistency, and reusability.

---

🔍 Exploratory Data Analysis (EDA)

Key analytical steps:

• Statistical summaries to detect outliers and skewness
• Global geospatial visualizations using latitude/longitude
• WHO-region-based comparisons
• Time-series trend analysis for confirmed, active, recovered, and death cases
• Correlation analysis between confirmed cases and mortality

Visualizations include:

• Global scatter maps
• Time-series line plots
• Regional bar charts
• Regression plots for severity analysis

---

🤖 Machine Learning Models

1️⃣ Mortality Prediction (Regression)

Target: Predict number of deaths

Models implemented:

• Random Forest Regressor
• Support Vector Machine (SVM)

Evaluation metrics:

• Mean Squared Error (MSE)
• R² Score
• Feature importance analysis

✔ Random Forest demonstrated strong predictive performance and interpretability
✔ Feature importance revealed active and confirmed cases as key mortality drivers

---

2️⃣ Patient Outcome Prediction (Classification)

Binary classification:

• Recovery-Dominant (1)
• Death-Dominant (0)

Model:

• Random Forest Classifier

Metrics:

• Precision
• Recall
• F1-Score
• Confusion Matrix

✔ Achieved high overall accuracy
✔ Strong performance in identifying recovery-dominant scenarios

---

📈 Key Insights

• Active case volume is a strong predictor of mortality
• Confirmed cases strongly correlate with deaths
• Recovery trends vary significantly by WHO region
• Pandemic severity peaks mid-timeline rather than uniformly
• Tree-based models outperform kernel-based methods for this dataset

---

🛠️ Tech Stack

• Python
• Pandas, NumPy
• Matplotlib, Seaborn
• Scikit-learn
• Jupyter Notebook
• Basemap (geospatial visualization)

---