H1N1 and Seasonal Flu Vaccines
Project links
Skills
About this project
H1N1 and Seasonal Flu Vaccine Prediction Using Machine Learning
Institution: University of Essex (MSc Coursework Project)
Tools & Technologies: Python, Pandas, NumPy, Scikit-learn, Imbalanced-learn, Matplotlib, Seaborn
This project focused on building machine learning models to predict individuals’ likelihood of receiving H1N1 and seasonal flu vaccines using demographic, behavioral, and health-related survey data. The dataset, sourced from the DrivenData “Flu Shot Learning” competition, contained 26,000+ records with 36 features covering demographics, medical history, and opinions about vaccines. Two binary classification tasks were performed:
Task 1: Predict whether an individual would receive the seasonal flu vaccine.
Task 2: Predict whether an individual would receive the H1N1 vaccine.
Data Preparation & Cleaning:
Combined multiple data files (features, labels, and test sets) into a structured format.
Addressed extensive missing values using KNN Imputer after encoding categorical variables.
Balanced the H1N1 target variable using SMOTE to improve model fairness.
Feature Engineering & Analysis:
Encoded categorical features (age group, education, income level, marital status, etc.).
Conducted hypothesis testing (Chi-square tests) to assess significance of features like education, poverty, and doctor’s recommendation on vaccination.
Performed exploratory data analysis (EDA) to uncover distributions, patterns, and correlations.
Model Development:
Implemented and compared multiple classifiers: Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, SVM, and Naive Bayes.
Applied cross-validation and grid search for tuning hyperparameters.
Separate models built for H1N1 and Seasonal Flu vaccines.
Model Evaluation:
Evaluated models using Accuracy, Precision, Recall, F1-Score, and AUC-ROC.
Seasonal flu prediction achieved ~86% accuracy with Gradient Boosting.
H1N1 vaccine prediction achieved ~92% accuracy with Gradient Boosting, outperforming other methods.
Designed robust machine learning pipelines capable of handling missing data and imbalanced classes.
Identified significant demographic and behavioral drivers behind vaccine adoption.
Demonstrated the superiority of ensemble models (Random Forest, Gradient Boosting) over traditional classifiers.
Delivered actionable insights for public health outreach strategies by targeting groups less likely to vaccinate.
Large proportion of missing values in health and income-related data.
Strong imbalance in H1N1 vaccine adoption rates.
Avoiding model overfitting while maintaining interpretability.
Capturing behavioral and attitudinal nuances through feature engineering.
Incorporate time-series survey data for trend-based predictions.
Extend modeling to multi-class classification (e.g., no vaccine, only H1N1, only seasonal, both).
Explore ensemble stacking and deep learning architectures to further boost accuracy.
Use interpretability frameworks (e.g., SHAP, LIME) for transparent decision-making in public health.