What technique involves creating many slightly different versions of the data by randomly resampling?

What technique involves creating many slightly different versions of the data by randomly resampling?

• A. Bootstrapping
• B. Random Forests
• C. Boosting
• D. Gradient Boosting

Answer: (A)

Bootstrapping centers on creating many slightly different datasets by sampling with replacement from the original data. Each bootstrap sample has the same size as the original but includes some observations more than once and leaves others out, so you end up with variety across samples. By calculating the statistic of interest on each bootstrap sample, you build an empirical distribution that reflects how that statistic would vary with repeated sampling, without needing strong parametric assumptions about the underlying data. This is especially useful in risk analytics for estimating confidence intervals around metrics like VaR or model results when the true distribution is unknown. The other techniques are more about how models are built: random forests use bootstrap samples plus random feature selection to train many trees, while boosting and gradient boosting sequentially improve models based on prior errors rather than focusing on creating multiple data versions.

Bootstrapping centers on creating many slightly different datasets by sampling with replacement from the original data. Each bootstrap sample has the same size as the original but includes some observations more than once and leaves others out, so you end up with variety across samples. By calculating the statistic of interest on each bootstrap sample, you build an empirical distribution that reflects how that statistic would vary with repeated sampling, without needing strong parametric assumptions about the underlying data. This is especially useful in risk analytics for estimating confidence intervals around metrics like VaR or model results when the true distribution is unknown. The other techniques are more about how models are built: random forests use bootstrap samples plus random feature selection to train many trees, while boosting and gradient boosting sequentially improve models based on prior errors rather than focusing on creating multiple data versions.