1. Project Understanding

• Objective
• Description

2. Data Understanding

• Import Libraries
• Load data

• 2.1. Data Cleaning (On how to get high-quality data?)

• Missing values (imputation, masking)
• Dealing with (duplication records, errors, inconsistency, misspelling, outliers etc)
• Categorical data: high and low cardinality

• 2.2. Exploratory Data Analysis

• Feature Engineering (binning, encoding)
• Statistical summaries and visualisations
• Features correlation

3. Predictive Modeling

• Classification models:

• Logistic Regression (LR),
• Support Vector Machine (SVM),
• K-nearest Neaighbors (Knn),
• Ensemble methods:

      . Random Forest (RF), Extra Trees (ET) 
      . Boosting: AdaBoost (AB), Extreme Gradient Boost (XGB)
      . Stacking: Decision Tree (DT) + AdaBoost (AB)

• Models Validation:

• GridSearch Cross-Validation

• Models Evaluation:

• Error metrics: Accuracy, Precision, Recall, F-score, Confusion Matrix

4. Conclusion

• Best model in terms of performance and runtime
• Alternative model
• Model Flaws

• Find the best Classification model and model hyperparameters to predict post-vaccine recovery (e.g. in terms of performance, time)

• To detect early warning AE signals and assess possible safety concerns for a given vaccine, which may generate hypotheses and prompt further investigations.

• what is VAERS?

  • Vaccine Adverse Event Reporting System --- A passive surveillance system that monitors vaccine safety in the U.S. beyond clinical trials.

• Who reports there?

  • Anyone can report to VAERS, including patients, parents, and healthcare providers. Healthcare providers are also required by law to do so. source: (https://vaers.hhs.gov/data/datasets.html)

• How we can make use of this data?

  • "VAERS reports alone cannot be used to determine if a vaccine caused or contributed to an adverse event or illness"(does not infer causality) source: (https://wonder.cdc.gov/wonder/help/vaers.html)

Levels of Analysis: Stats, Visualization

• At a glance: categorical values || missing values; what is the best imputation method?

• Feautres Distribution

  • Post-covid19-vaccination Deaths by

  • Vaccination doses counts in 2020 and 2021.
  • Vaccine manufacture and vaccine lots.
  • Recovery counts by each State.

  • Gender, ageGroup, vaccine brand and vaccine lot distribution in

  • Hospitalization, Emergency visit, life threatening, disability, prolonged hospitalized days
  • Myocarditis/Pericarditis, Guillain–Barré Syndrome, Thrombosis with Thrombocytopenia Syndrome (TTS) (Rare but high risk health conditions attributed to the vax administration.)

__Q. Does the population-health status play a significant role in having post-vaccine AE or getting recoverd?

• status of the vaccine recipients: (ageGroup, gender, medical history, current illnes, current medication, allergy)

  with:

• Recovery outcomes: (yes, no) after Vaccine administration
• Serious outcomes: (yes, no) after Vaccine administration (Death, life threatening, disability, emergency visit, prolonged hospitalized days)

4. Conclusion

• RF classifier was proposed as the best model in terms of performance and runtime
• XGB as an alternative model

• Model Flaws:

• Create different featured engineered variables from independent variables, that may improve correlation with the dependent variable and see if it helps in improving model performance and feature importance.
• Keep iterating with subsample, learning rate and estimators for the `stacking model to further improve the results
• Explore the database updates, more data will definitely help in improving the prediction outcome.