Master To Do

[egmavis]

Data Exploration

• show initial balance of gender and race of raw data

Pre-processing

• create dataframes
• create different balanced/imbalanced data sets
• combine indian and asian labels into one "asian" label, adjust "other" label accordingly

Modeling (on all data split using train/val/test)

• Logistic Regression (adjusting regularization hyperparameter)
• Random Forest (more advanced baseline model, start with default parameters)
• Tensorflow NN (additional workflow required)

Select best performing model for running the following experiments (will prob be the NN):

Experiments

• Leave-one-out racial subgroups (white, black, asian, other)
• final chosen model (if too slow, random forest)
• Skewed gender balance (80% male 20% female, and 20% male 80% female)
• final chosen model (if too slow, random forest)

Visualizations

• ROC/PR curves for each model (data as a whole and for subgroup experiments)
• F1 scores

Final Reporting

• Follow rubric to confirm all expectations are met

Presentation Slides

Notes

• we will use all data instead of random sample
• each modeling process overall and for each experiment needs to be redone so as to avoid data snooping

[nickeubank]

Look at you using issues and checklists for Kyle's class too! Yay!!