Sushmita V Gopalan - Predicting Infant Mortality

[rickecon]

Killed it. Really good.

[bensoltoff]

• Look into marginal effects plots for logistic regression model coefficients. Helps improve interpretability
• Never heard of Tomek Links before. Interesting approach. And it doesn't substantially bias the results by excluding this specific subsample?
• Really nice research question. Good crossover between social scientific research with real-world implications and need

[jamesallenevans]

Infant mortality across countries Predict this in Ethiopia Removal of Tomek links, nearest neighbor of different class Very interesting...what are your theoretical interpretations about exactly what is driving this improvement.

[ghost]

Can't you just predict all infant as dying to minimize false negative? What are some constraints?

[jgdenby]

Can removing Tomek Links induce illegitimately distinct boundaries between groups? What if the noise is a feature of the population, not a bug resulting from the particular sampling?

[jmausolf]

I may have missed this in the presentation, but if cross-validation is a next step, were any of the data time-dependent and any of the false negative errors temporal?

[rodrigovaldes]

Great presentation. I would like a lager introduction on why it's important to reduce the false negatives. I mean, you introduced the general topic very well, but not the false negatives.