Project Description

We were interested in understanding peoples’ attitudes towards H1N1 and Seasonal Flu vaccination and predicting whether they would receive each, given such an examination’s possible relevance to the current 59% COVID-19 vaccination rate in the US. We performed this study in the context of a data science contest hosted by DrivenData.org. Their supplied datasets contained 52,414 survey observations, with 37 predictors and 2 response variables. We studied and feature engineered the training data before building five classification models. We evaluated our models’ performance locally and then submitted each to the competition’s auto-grader for neutral evaluation. Our Random Forest model performed the best and is at 86th percentile of all current submissions. We found that the strongest predictor of both H1N1 and Seasonal Flu vaccination acceptance was the doctor’s recommendation.

Data

We obtained our data from an ongoing competition hosted by DrivenData.org . They, in turn, obtained their raw data from the United States National Center for Health Statistics at the CDC and prepared it (using undisclosed methods) for the contest. The data consists of responses to a phone survey of 52,414 individuals which asked them whether or not they had received their H1N1 and Seasonal Flu vaccines (the outcomes we are trying to understand and predict) in conjunction with 35 other questions all of which fell into three categories:

1. Behavioural questions. E.g. Wears face-mask, avoids large gatherings, washes hands, touches face
2. Opinion surveys. E.g. Is concerned about the vaccine, is knowledgeable about vaccine
3. Individual attributes E.g. Age, sex, race, marital status, income, occupation

Models

We sought to create two binary classifiers that would yield yes/no predictions for whether an individual received the H1N1 and Seasonal Flu vaccines based on the 35 features in the Training Set. We selected some popular classification techniques/algorithms and fine-tuned them based on their performance on the Evaluation Set.

Conclusions

In a descriptive context, and for both H1N1 and Seasonal Flu vaccination, we observed that doctor recommendation is significantly more important than other factors. Using Regularization techniques, we found that an individual is 5.3 times more likely to take the H1N1 vaccine and 4.4 times more likely to take the Seasonal Flu vaccine if their doctor recommends it to them than if their doctor did not.