A data-driven approach to the forecasting of ground-level ozone concentration.
Citations
8 citations
2 citations
References
40,785 citations
"A data-driven approach to the forec..." refers methods in this paper
...Random Forests and Quantile Random Forests The random forest (RF) algorithm independently fits several decision trees, each trained on different datasets, created from the original one through random re-sampling of the observations and keeping only a fraction of the overall features, chosen at random (Hastie, Tibshirani and Friedman, 2009). The final prediction of the RF is then a (possibly weighted) average of the trees’ responses. One important variant of RF algorithms are Quantile Regression Forests (QRF); the main difference from RF is that QRF keeps the value of all the observations in the fitted trees’ nodes, not just their mean, and assesses the conditional distribution based on this information. In this paper, we have used theMatlab TreeBagger class, which implements the QRF algorithm described in Meinshausen (2014). Tree-based boosting algorithms Boosting algorithms employ additive training: starting from a constant model, at each iteration a new tree or any other so called "weak learner" hk(x) is added to the overall model Fk(x), so that Fk+1(x) = Fk(x) + hk(x) where ≤ 1 is a hyper-parameter denoting the learning rate, which helps reducing over-fitting....
[...]
...Instead of punishing using the L2 norm, the LASSO (Least Absolute Shrinkage and Selection Operator) regression (Tibshirani, 1996) penalizes using the L1 norm, such that some of the elements of ̂ could be set to zero....
[...]
14,872 citations
11,507 citations
10,306 citations
4,644 citations