Mortality Trends Analysis Advanced Data Analysis | EDA | Time-Series Modeling | Forecasting
This project analyzes weekly mortality data from the United States using advanced data-science techniques including exploratory data analysis (EDA), feature engineering, trend decomposition, and machine-learning-based forecasting.
The goal is to identify major mortality patterns, understand cause-wise trends, and build predictive models to forecast all-cause mortality.
Project Highlights
Performed comprehensive EDA to understand mortality behavior
Built feature-engineered time-series dataset
Applied linear regression forecasting using lag & moving-average features
Conducted trend + seasonality decomposition
Created heatmaps, time-series plots, and stacked area charts
Extracted meaningful insights such as peak mortality weeks and correlations
Dataset
Source: Weekly mortality records (Excel format) containing:
All-Cause deaths
COVID-19 deaths
Major causes (heart disease, cancer, etc.)
Date of week ending
Jurisdiction (state names)
Technologies Used
Python 3
Pandas, NumPy
Matplotlib, Seaborn
Statsmodels (seasonal decomposition)
Scikit-learn (machine learning)
Project Workflow
- Import & Clean Data
Loaded Excel dataset
Converted date fields
Filtered U.S-level aggregation
Filled missing values
Sorted chronologically
- Feature Engineering
Engineered several predictive & analytical features:
Lag variables (1-week lag)
Moving Averages (4-week average)
Week number
Year
Rolling trend signals
- Exploratory Data Analysis
Includes:
Time-series trend plots
Correlation matrix
Distribution of causes
Stacked cause-wise area charts
Jurisdiction heatmap
- Time-Series Trend Decomposition
Using seasonal_decompose():
Trend
Seasonal patterns
Residual noise
- Machine Learning Forecast
Built a Linear Regression model using:
AllCause_Lag1
COVID_Lag1
AllCause_MA4
COVID_MA4
Evaluated using MAE & R², with a prediction visualization.
- Results & Key Insights Peak Mortality
Highest COVID-19 mortality week identified
Highest All-Cause mortality week identified
Correlation
COVID-19 and All-Cause Mortality correlation example:
Strong positive correlation → pandemic impact visible in overall mortality.
Year-wise mortality totals
Helps track increase or decrease over years.
Visualizations Included
All-Cause vs Natural Deaths Trend
Correlation Heatmap
Time-Series Decomposition
Prediction vs Actual Line Plot
Stacked Area Chart for top causes
Jurisdiction Heatmap