COVID-19 ICU Prediction Analysis 🏥📊

Overview

This repository contains a Data Science assessment project completed for the GP8R46 NPA Data Science course. The goal of this project is to act as a Data Scientist for the Scottish Government to assist in the COVID-19 pandemic response.

By analyzing daily published statistics, this project aims to predict the number of patients requiring Intensive Care Units (ICU), enabling the NHS to allocate resources effectively.

📂 Dataset

The analysis uses a distilled version of open license relational data from the Scottish Government.

• Data Source: covid19.csv
• Key Features: Dates, First/Second Doses, Hospital Admissions, Positive Tests, and ICU numbers.

🛠️ Technologies Used

• Python 🐍
• Pandas (Data Manipulation)
• NumPy (Numerical Analysis)
• Matplotlib & Seaborn (Data Visualization)
• Scikit-Learn (Machine Learning)

🔍 Key Analysis Steps

1. Data Cleaning: Handling missing values and structuring the dataset for analysis.
2. Exploratory Data Analysis (EDA):
   • Statistical summary of the data.
   • Visualizing relationships between variables (e.g., Second Dose vs. ICU, Positive Tests vs. ICU).
   • Correlation analysis to identify key predictors.
3. Feature Selection: Identified positive_tests as the feature most strongly correlated with ICU admissions.
4. Machine Learning:
   • Splitting data into Training (90%) and Testing (10%) sets.
   • Training a Linear Regression model.
   • Evaluating model performance using $R^2$ scores.

📈 Model Performance

The project compares two modeling approaches to predict ICU numbers:

1. Linear Regression

• Training Score: ~91.4%
• Testing Score: ~87.8%
• Observation: The model provides a decent baseline but struggles with complex patterns.

2. Random Forest (Comparison)

• Training Score: ~99.0%
• Testing Score: ~94.7%
• Conclusion: The Random Forest model outperformed Linear Regression. It was able to capture non-linear relationships in the data (e.g., high vaccination rates dampening the effect of positive cases on ICU admissions).

🚀 How to Run

1. Clone this repository.
2. Ensure you have the required libraries installed:

   pip install pandas numpy matplotlib seaborn scikit-learn

3. Open the Jupyter Notebook Covid19_ICU_Prediction_Analysis.ipynb to view the analysis and code.

📜 Assessment Context

This evidence was produced for the Combined J2G246 Data Science, J2HN46 Data Citizenship & J2G646 Machine Learning Assessment.