US 11,875,878 B2
Machine learning method and apparatus using steps feature selection based on genetic algorithm
Seung Wan Kang, Seoul (KR); Namheon Kim, Paju-si (KR); and Dong Won Yang, Seoul (KR)
Assigned to IMEDISYNC. LTD., Seoul (KR); and THE CATHOLIC UNIVERSITY OF KOREA INDUSTRY-ACADEMIC COOPERATION FOUNDATION, Seoul (KR)
Appl. No. 17/624,500
Filed by IMEDISYNC. LTD., Seoul (KR); and THE CATHOLIC UNIVERSITY OF KOREA INDUSTRY-ACADEMIC COOPERATION FOUNDATION, Seoul (KR)
PCT Filed Nov. 8, 2021, PCT No. PCT/KR2021/016145
§ 371(c)(1), (2) Date Jan. 3, 2022,
PCT Pub. No. WO2022/139168, PCT Pub. Date Jun. 30, 2022.
Claims priority of application No. 10-2020-0180089 (KR), filed on Dec. 21, 2020.
Prior Publication US 2022/0399081 A1, Dec. 15, 2022
Int. Cl. G01N 33/48 (2006.01); G01N 33/50 (2006.01); G16B 40/00 (2019.01); G16B 5/00 (2019.01); G06N 3/126 (2023.01)
CPC G16B 40/00 (2019.02) [G06N 3/126 (2013.01); G16B 5/00 (2019.02)] 15 Claims
 
1. A machine learning method using steps feature selection based on a genetic algorithm, comprising:
defining a feature set including a plurality of features, wherein the defining of the feature set includes defining each of the plurality of features as a combination between an electroencephalogram occurrence location and an electroencephalogram frequency band for an electroencephalogram signal;
generating a plurality of feature combinations including n-dimensional features (n is a natural number) for the feature set;
independently constructing feature models for the plurality of feature combinations and calculating prediction accuracy for each of the feature models as a prediction result for a predetermined data set;
arranging the feature models according to the prediction accuracy to determine at least one feature model that satisfies a preset criterion;
determining a first feature from among features included in a corresponding feature set of the at least one feature model;
updating the feature set to include only the first feature and re-determining a feature model for a (n+1)-dimensional feature combination based on the updated feature set, and
diagnosing one or more pathological symptoms using the feature model for the (n+1)-dimensional feature combination,
wherein the calculating of each of the prediction accuracy includes calculating the prediction accuracy based on whether each of the feature models matches a predicted value and an actual value regarding the presence or absence of amyloid for the predetermined data set.