Project Details
Description
The general aim of this project is to develop a comprehensive analytic framework to quantify and estimate predictability and explainability of a single random variable Y using the information contained in a set of potential explanatory random variables X_1, …, X_d -including in particular the situation with d being large.
Our approach enables a dimension reduction by converting the original (d+1)-dimensional problem to the bivariate setting, with the crucuial point that the underlying transformation preserves the key information about the dependence between Y and the random variables X_1,...,X_d. This not only enables the quantification of predictability and explainability using known bivariate tools, but also allows for a fast, precise and efficient estimation.
The proposed approach is truly innovative and a very promising contributions with a broad range of applicability. We expect our methods to be highly recognised in practice.
Our approach enables a dimension reduction by converting the original (d+1)-dimensional problem to the bivariate setting, with the crucuial point that the underlying transformation preserves the key information about the dependence between Y and the random variables X_1,...,X_d. This not only enables the quantification of predictability and explainability using known bivariate tools, but also allows for a fast, precise and efficient estimation.
The proposed approach is truly innovative and a very promising contributions with a broad range of applicability. We expect our methods to be highly recognised in practice.
| Short title | ReDim |
|---|---|
| Acronym | ReDim |
| Status | Active |
| Effective start/end date | 1/10/22 → 28/02/26 |