A Genetic Programming-Driven 1 q) k6 V+ Z6 W) b
Data Fitting Method 2 G7 w5 i9 F1 i/ U$ y9 G
$ }( a) l$ }. D9 \8 _+ v4 y4 u
* o' r8 g k6 Q& Q* X2 ?4 R S; v4 ?8 B1 t* c/ A
Data fifitting is the process of constructing a curve, or a set of mathematical functions, that has& H' z- Y, a1 Q8 W
the best fifit to a series of data points. Different with constructing a fifitting model from same type of function,
. G" v1 t: O2 L, Dsuch as the polynomial model, we notice that a hybrid fifitting model with multiple types of function may have% t1 {. S3 q; B( s) g1 F7 g0 m. u
a better fifitting result. Moreover, this also shows better interpretability. However, a perfect smooth hybrid
* E% y0 Y) M/ a- ~/ M6 hfifitting model depends on a reasonable combination of multiple functions and a set of effective parameters.
4 T3 G6 }; G4 C9 I/ |% k! n! k( P! sThat is a high-dimensional multi-objective optimization problem. This paper proposes a novel data fifitting
( k+ U3 S( T" t; S& Nmodel construction approach. In this approach, the model is expressed by an improved tree coding expression" ~, A) K4 z! p, X
and constructed through an evolution search process driven by the genetic programming. In order to verify/ q; Z$ a, F! O
the validity of generated hybrid fifitting model, 6 prediction problems are chosen for experiment studies. The
j& f, y( e5 P4 j# n( aexperimental results show that the proposed method is superior to 7 typical methods in terms of the prediction( b+ O% G* b* {4 b
accuracy and interpretability
/ S) }% m- B& [% r1 M. K7 r1 E# z2 O- L o
' A- ~9 c, b4 _# T Q: |
& N5 ~% t4 R1 x8 W7 I
/ s7 a* H& A' p! A$ W! }
2 {1 i5 l2 D% p+ n6 f# X7 s- V" N
|