US 12,437,198 B2
Classification of a non-monetary donation based on machine learning
Thorsten Muehge, Budenheim (DE); Erik Rueger, Ockenheim (DE); Markus Ettl, Yorktown Heights, NY (US); Beverley Joanne Dyke, London (GB); and Thomas William Moncreaff, Emsworth (GB)
Assigned to INTERNATIONAL BUSINESS MACHINES CORPORATION, Armonk, NY (US)
Filed by INTERNATIONAL BUSINESS MACHINES CORPORATION, Armonk, NY (US)
Filed on Jul. 22, 2020, as Appl. No. 16/935,538.
Prior Publication US 2022/0027682 A1, Jan. 27, 2022
Int. Cl. G06N 3/082 (2023.01); G06F 18/20 (2023.01); G06F 18/21 (2023.01); G06F 18/2113 (2023.01); G06F 18/2415 (2023.01); G06N 3/045 (2023.01); G06N 3/08 (2023.01); G06N 20/00 (2019.01)
CPC G06N 3/082 (2013.01) [G06F 18/2113 (2023.01); G06F 18/217 (2023.01); G06F 18/2415 (2023.01); G06F 18/285 (2023.01); G06N 20/00 (2019.01); G06N 3/08 (2013.01)] 6 Claims
OG exemplary drawing
 
1. A method for classifying a non-monetary donation, the method comprising:
training a first machine learning model using supervised machine learning, the trained first machine learning model comprising input neurons arranged in an input layer, each input neuron being operable for inputting a value (X) of a corresponding input parameter of the non-monetary donation, and at least one hidden layer comprising multiple hidden neurons, each hidden neuron being operable for calculating a hidden layer value based on at least one value of the corresponding input parameter and first weights associated with respective input neurons, the trained first machine learning model having a total number of input parameters and a total number of hidden neurons;
generating a matrix associated with model weights, wherein rows of the matrix correspond to the respective input neurons and columns of the matrix correspond to respective hidden neurons;
calculating a row total for each row of the matrix by summing a first absolute value of each first weight in each row;
calculating a column total for each column by summing a second absolute value of each second weight in each column;
selecting, from the trained first machine learning model having the total number of input parameters and the total number of hidden neurons, a reduced number of input parameters based on ranking the row totals in the matrix and a reduced number of hidden neurons based on ranking the column totals in the matrix;
determining whether to accept the non-monetary donation or reject the non-monetary donation based on a classification of the non-monetary donation using the trained first machine learning model having the total number of input parameters and the total number of hidden neurons, wherein rejecting the non-monetary donation includes refusing to accept the non-monetary donation for any donation recipient;
in response to the classification, using the trained first machine learning model having the total number of input parameters and the total number of hidden neurons, being an uncertain classification as to accepting the non-monetary donation or rejecting the non-monetary donation, verifying the uncertain classification of the non-monetary donation by processing the selected reduced number of input parameters without processing any unselected input parameters using the trained first machine learning model having the selected reduced number of hidden neurons and excluding any unselected hidden neurons to obtain a second classification; and
retraining the trained first machine learning model having the total number of input parameters and the total number of hidden neurons using the second classification.