TY - GEN
T1 - Effective pre-processing of genetic programming for solving symbolic regression in equation extraction
AU - Koga, Issei
AU - Ono, Kenji
N1 - Publisher Copyright:
© Springer Nature Switzerland AG 2019.
PY - 2019
Y1 - 2019
N2 - Estimating a form of equation that explains data is very useful to understand various physical, chemical, social, and biological phenomena. One effective approach for finding the form of an equation is to solve the symbolic regression problem using genetic programming (GP). However, this approach requires a long computation time because of the explosion of the number of combinations of candidate functions that are used as elements to construct equations. In the present paper, a novel method to effectively eliminate unnecessary functions from an initial set of functions using a deep neural network was proposed to reduce the number of computations of GP. Moreover, a method was proposed to improve the accuracy of the classification using eigenvalues when classifying whether functions are required for symbolic regression. Experiment results showed that the proposed method can successfully classify functions with over 90% of the data created in the present study.
AB - Estimating a form of equation that explains data is very useful to understand various physical, chemical, social, and biological phenomena. One effective approach for finding the form of an equation is to solve the symbolic regression problem using genetic programming (GP). However, this approach requires a long computation time because of the explosion of the number of combinations of candidate functions that are used as elements to construct equations. In the present paper, a novel method to effectively eliminate unnecessary functions from an initial set of functions using a deep neural network was proposed to reduce the number of computations of GP. Moreover, a method was proposed to improve the accuracy of the classification using eigenvalues when classifying whether functions are required for symbolic regression. Experiment results showed that the proposed method can successfully classify functions with over 90% of the data created in the present study.
UR - http://www.scopus.com/inward/record.url?scp=85072858902&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85072858902&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-30284-9_6
DO - 10.1007/978-3-030-30284-9_6
M3 - Conference contribution
AN - SCOPUS:85072858902
SN - 9783030302832
T3 - Communications in Computer and Information Science
SP - 89
EP - 103
BT - Information Search, Integration, and Personalization - 12th International Workshop, ISIP 2018, Revised Selected Papers
A2 - Kotzinos, Dimitris
A2 - Laurent, Dominique
A2 - Spyratos, Nicolas
A2 - Tanaka, Yuzuru
A2 - Taniguchi, Rin-ichiro
PB - Springer Verlag
T2 - 12th International Workshop on Information Search, Integration and Personalization, ISIP 2018
Y2 - 14 May 2018 through 15 May 2018
ER -