US 12,248,534 B2
Automated feature engineering for predictive modeling using deep reinforcement learning
Mohamed Bouadi, La Garenne-Colombes (FR); and Arta Alavi, Andresy (FR)
Assigned to SAP SE, Walldorf (DE)
Filed by SAP SE, Walldorf (DE)
Filed on Mar. 14, 2022, as Appl. No. 17/694,288.
Claims priority of provisional application 63/291,789, filed on Dec. 20, 2021.
Prior Publication US 2023/0195842 A1, Jun. 22, 2023
Int. Cl. G06N 3/08 (2023.01); G06F 18/21 (2023.01); G06F 18/214 (2023.01); G06N 3/006 (2023.01); G06N 3/084 (2023.01); G06N 3/092 (2023.01); G06N 5/04 (2023.01); G06N 7/01 (2023.01); G06N 20/00 (2019.01)
CPC G06F 18/217 (2023.01) [G06F 18/214 (2023.01); G06N 3/08 (2013.01)] 18 Claims
OG exemplary drawing
 
7. A method comprising:
receiving a dataset comprising a plurality of features, each of the plurality of features associated with a plurality of values;
responsive to receiving the dataset, and for each of the plurality of features, determining a statistical measure from the plurality of values associated with the feature;
inputting the statistical measures into a neural network;
receiving, from the neural network, a selection of an action;
responsive to receiving the selection of the action, transforming the dataset by applying the action to at least one of the plurality of features to obtain a transformed dataset including a second plurality of features;
calculating a reward for the neural network based on the action;
determining if a budget for the neural network is reached;
responsive to determining the budget is reached, storing the plurality of features in a final model; and
responsive to determining the budget is not reached, determining a second statistical measure of each of the second plurality of features from the second plurality of values associated with each of the second plurality of features and inputting the second statistical measures and the reward into the neural network.