R2 Metric Dynamics for Decesion Trees Regression Models Trained on Series of Different Sizes
Yurii Babich · 2024
Machine learning results presented in this paper clearly display that R2score can vary significantly depending on the samples selected to test part of a series used for regression model evaluation. The mentioned variation can contribute to model's accuracy overestimation. The R2score dynamics is examined using 28000 cycles of a Decision Tree regression model training and evaluation using series of different sizes. The paper both states and proves a hypothesis that the R2score variation is expected to increase with series size reduction and the variation is supposed to be observed for models trained on the same series because of training/test samples selection randomness. The experiments carried out allowed to propose an alternative approach that did not require any supplementary metrics (including Mean Square Error, Mean Absolute Error or Mean Error) to be involved in a model estimation. The proposed approach considers application of the R2score along with its variation that must not exceed 0.2 for the Decision Tree regression model.