COVID Vaccine Prediction — Group S

Research Methods in Data Science | University of Hertfordshire

Data Science Challenge Hackathon

---

Overview

This repository contains the full work produced by Group S during the Data Science Challenge Hackathon. The goal was to build a machine learning pipeline to predict COVID-19 vaccine uptake from a survey dataset, submitted as part of the group project assessment.

Core modelling and EDA was completed during the hackathon session on 5th June 2026. Explainable AI (SHAP analysis) and figure polishing were completed afterwards as permitted by the assignment brief.

---

Contributors

Name	GitHub
Akhila Thurpati	@Akhila152003
Bhavin Thakur	@bhavinthakur29
Jyothi Kallubhavi	@Kallubhavijyothi
Sriharshini Thatiparthi	@sriharshini1603-sketch
Upender Madha	@upender8096
Venkat Bhargav Inkollu	@InkolluvenkatBhargav

---

Repository Structure

├── challenge_submission_group_S.ipynb   # Main notebook (all code)
├── dataset_C_testing.csv                # Testing dataset
├── dataset_C_training.csv               # Training dataset
├── plots/                               # All saved visualisation outputs
│   ├── target_distribution.png
│   ├── target_balance.png
│   ├── missing_values.png
│   ├── correlation_heatmap.png
│   ├── top_correlations.png
│   ├── behaviour_vs_vaccination.png
│   ├── pca_projection.png
│   ├── best_model_performance.png
│   ├── permutation_importance.png
│   ├── model_comparison.png
│   ├── decision_threshold.png
│   ├── shap_bar.png
│   ├── shap_waterfall.png
│   └── learning_curve.png
├── submission_csv/                      # All CSV predictions for evaluation
│   ├── challenge_submission_group_S_order_1.csv
│   ├── challenge_submission_group_S_order_2.csv
│   ├── challenge_submission_group_S_order_3.csv
│   ├── challenge_submission_group_S_order_4.csv
│   ├── challenge_submission_group_S_order_5.csv
│   └── FINAL_Group_S_Submission.csv
└── README.md

---

Dataset

Dataset: Hackathon Dataset C (dataset_C_training.csv, dataset_C_testing.csv)
Source: Hackathon Dataset C Information.pdf

The dataset contains 31 columns — a unique respondent ID, 29 survey-based features covering behavioural habits, medical conditions, opinions on COVID risk and vaccine effectiveness, and demographic information, with covid_vaccine as the binary target variable.

---

Notebook Structure

Section	Description
1	Imports & constants
2	Load data
3	Exploratory Data Analysis (EDA)
4	Preprocessing & Feature Engineering
5	Cross-Validation setup (StratifiedKFold)
6	ML Models — Logistic Regression, KNN, Random Forest, Neural Network, Gradient Boosting
7	Performance Metrics — confusion matrix, ROC curve, precision-recall curve
8	Explainable AI — Permutation Importance
9	Hyperparameter Tuning — GridSearchCV
9.1	Model Comparison — all 6 models across F1, ROC-AUC, Accuracy
10	Decision-Threshold Optimisation
11	XAI — SHAP Horizontal Bar Chart (completed post-hackathon)
12	XAI — SHAP Waterfall Plot (completed post-hackathon)
13	Learning Curve (Bias/Variance analysis)
14	Final Submissions — ranked by F1 score
15	Critical Analysis

---

Models Trained

• Logistic Regression
• K-Nearest Neighbors (KNN)
• Random Forest
• Neural Network (MLP)
• Gradient Boosting (HistGradientBoosting)
• Tuned Gradient Boosting (GridSearchCV optimised)

---

Key Results

Model	ROC-AUC	Accuracy	F1
GradientBoosting	0.8566	0.8130	0.6927
TunedGradientBoosting	0.8560	0.8110	0.6890
LogisticRegression	0.8340	0.7940	0.6450
NeuralNetwork	0.8200	0.7840	0.6330
RandomForest	0.8340	0.7840	0.6160
KNN	0.8060	0.7730	0.5680

All scores are 5-fold stratified cross-validation results on the training set.
Submissions are ordered by F1 score as per evaluation criteria.

---

Dependencies

All standard packages are pre-installed in Google Colab. The notebook auto-installs any missing packages on first run:

numpy, pandas, matplotlib, scikit-learn
lightgbm
shap

---

How to Run

1. Clone this repository
2. Open challenge_submission_group_S.ipynb in Google Colab or Jupyter
3. Run all cells top to bottom
4. Predictions will be saved to submission_csv/ and plots to plots/

---

Collaboration

All code was developed collaboratively using a shared Google Colab notebook linked to this GitHub organisation repository. Commit history and Colab activity logs serve as evidence of individual contributions during the hackathon.

---

University of Hertfordshire — Research Methods in Data Science — June 2026