D. Serrano Ortega, E. García Portugués
A new methodology to quantify the uncertainty in a random forest prediction is presented for the general case where the response and predictors are defined in metric spaces. The confidence regions utilize out-of-bag observations generated during a single forest training. In this way, the entire dataset is used for both prediction and uncertainty estimation, which results in computationally efficient estimations. Asymptotic coverage theory is presented in four different scenarios for the type of coverage. The proposed prediction regions are illustrated in different metric spaces.
Keywords: Confidence region; Metric spaces; Fréchet regression; Random forests; Out-of-bag errors
Scheduled
Nonparametric Statistics: High Dimension
June 12, 2025 5:10 PM
Sala de prensa (MR 13)