J. C. Castro Gómez, E. Carrizosa Priego, V. Guerrero
In many real-world scenarios, data are collected from diverse sources, some of which can be costly to access in terms of resources or time. This often leads to incomplete datasets in the sense that blocks of features are available only for some of the observations, thus yielding a block-missing data set. Rather than discarding or imputing missing data, in this work we propose a greedy tree-based method that makes use of all available data to train a predictive model.
At each step, this greedy algorithm evaluates the option of making a split based on a block of features, aiming to balance predictive accuracy with the cost of accessing additional information. Designed for both classification and regression tasks, our approach integrates an adaptive decision-making mechanism to optimize the trade-off between data completeness and prediction performance.
Palabras clave: Pescriptive model, Classification and Regression Trees, Random Forest
Programado
GT03. AMC1 Aprendizaje Automático
10 de junio de 2025 11:30
Auditorio 1. Ricard Vinyes