Imbalanced data classification has been a major topic in the machine learning community. Different approaches can be taken to solve the issue in recent years, and researchers have given a lot of attention to data level techniques and algorithm level. However, existing methods often generate samples in specific regions without considering the complexity of imbalanced distributions. This can lead to learning models overemphasizing certain difficult factors in the minority data. In this paper, a Monte Carlo sampling algorithm based on Gaussian Mixture Model (MCS-GMM) is proposed. In MCS-GMM, we utilize the Gaussian mixed model to fit the distribution of the imbalanced data and apply the Monte Carlo algorithm to generate new data. Then, in order to reduce the impact of data overlap, the three sigma rule is used to divide data into four types, and the weight of each minority class instance based on its neighbor and probability density function. Based on experiments conducted on Knowledge Extraction based on Evolutionary Learning datasets, our method has been proven to be effective and outperforms existing approaches such as Synthetic Minority Over-sampling TEchnique.
Citation: Gang Chen, Binjie Hou, Tiangang Lei. A new Monte Carlo sampling method based on Gaussian Mixture Model for imbalanced data classification[J]. Mathematical Biosciences and Engineering, 2023, 20(10): 17866-17885. doi: 10.3934/mbe.2023794
[1] |
Linglong Du, Min Yang .
Pointwise long time behavior for the mixed damped nonlinear wave equation in |
[2] |
Linglong Du .
Long time behavior for the visco-elastic damped wave equation in |
[3] | Hantaek Bae . On the local and global existence of the Hall equations with fractional Laplacian and related equations. Networks and Heterogeneous Media, 2022, 17(4): 645-663. doi: 10.3934/nhm.2022021 |
[4] | Günter Leugering, Sergei A. Nazarov, Jari Taskinen . The band-gap structure of the spectrum in a periodic medium of masonry type. Networks and Heterogeneous Media, 2020, 15(4): 555-580. doi: 10.3934/nhm.2020014 |
[5] | Seung-Yeal Ha, Gyuyoung Hwang, Hansol Park . Emergent behaviors of Lohe Hermitian sphere particles under time-delayed interactions. Networks and Heterogeneous Media, 2021, 16(3): 459-492. doi: 10.3934/nhm.2021013 |
[6] | Arianna Giunti . Convergence rates for the homogenization of the Poisson problem in randomly perforated domains. Networks and Heterogeneous Media, 2021, 16(3): 341-375. doi: 10.3934/nhm.2021009 |
[7] | Francesca Alessio, Piero Montecchiari, Andrea Sfecci . Saddle solutions for a class of systems of periodic and reversible semilinear elliptic equations. Networks and Heterogeneous Media, 2019, 14(3): 567-587. doi: 10.3934/nhm.2019022 |
[8] | Ciro D'Apice, Olha P. Kupenko, Rosanna Manzo . On boundary optimal control problem for an arterial system: First-order optimality conditions. Networks and Heterogeneous Media, 2018, 13(4): 585-607. doi: 10.3934/nhm.2018027 |
[9] | María Anguiano, Francisco Javier Suárez-Grau . Newtonian fluid flow in a thin porous medium with non-homogeneous slip boundary conditions. Networks and Heterogeneous Media, 2019, 14(2): 289-316. doi: 10.3934/nhm.2019012 |
[10] | Wen Shen . Traveling waves for conservation laws with nonlocal flux for traffic flow on rough roads. Networks and Heterogeneous Media, 2019, 14(4): 709-732. doi: 10.3934/nhm.2019028 |
Imbalanced data classification has been a major topic in the machine learning community. Different approaches can be taken to solve the issue in recent years, and researchers have given a lot of attention to data level techniques and algorithm level. However, existing methods often generate samples in specific regions without considering the complexity of imbalanced distributions. This can lead to learning models overemphasizing certain difficult factors in the minority data. In this paper, a Monte Carlo sampling algorithm based on Gaussian Mixture Model (MCS-GMM) is proposed. In MCS-GMM, we utilize the Gaussian mixed model to fit the distribution of the imbalanced data and apply the Monte Carlo algorithm to generate new data. Then, in order to reduce the impact of data overlap, the three sigma rule is used to divide data into four types, and the weight of each minority class instance based on its neighbor and probability density function. Based on experiments conducted on Knowledge Extraction based on Evolutionary Learning datasets, our method has been proven to be effective and outperforms existing approaches such as Synthetic Minority Over-sampling TEchnique.
In this paper, we study the pointwise long time behavior of the solution for the nonlinear wave equation with frictional and visco-elastic damping terms
{∂2tu−c2Δu+ν1∂tu−ν2∂tΔu=f(u),u|t=0=u0(x),ut|t=0=u1(x), | (1) |
in multi-dimensional half space
(a1∂x1u+a2u)(x1=0,x′,t)=0. | (2) |
Over the past few decades, many mathematicians have concentrated on solving different kinds of damped nonlinear wave equations. The first kind is called the frictional damped wave equation, which is given as follows
{∂2tu−c2Δu+ν∂tu=f(u),u|t=0=u0(x),ut|t=0=u1(x), | (3) |
see [9,19,20,23] for the references. It is showed that for the long time, the fundamental solution for the linear system of (3) behaves like the Gauss kernel
{∂2tu−c2Δu−ν∂tΔu=f(u),u|t=0=u0(x),ut|t=0=u1(x). | (4) |
One can refer to [22] for the decaying rate of the linear solution, [11,12] for the asymptotic profiles of the linear problem, [4,21] for the nonlinear equation, etc. In [9], the authors studied the fundamental solution for the linear system of (4). The results show that the hyperbolic wave transport mechanism and the visco-elastic damped mechanism interact with each other so that the solution behaves like the convented heat kernel, i.e.,
For the initial-boundary value problem of the different damped wave equations, many authors studied the global well-posedness existence, long time behaviors, global attractors and decaying rate estimates of some elementary wave by using delicate energy estimate method, for example [1,13,25,26,28,29]. In this paper, we will use the pointwise estimate technique to give the long time behavior of the solution for system (1) with boundary condition (2). The main part of this technique is the construction and estimation of the Green's functions for the following linear systems:
{∂2tG1−c2ΔG1+ν1∂tG1−ν2∂tΔG1=0,x1,y1>0,x′∈Rn−1,t>0,G1(x1,x′,0;y1)=δ(x1−y1)δ(x′),G1t(x1,x′,0;y1)=0,a1∂x1G1(0,x′,t;y1)+a2G1(0,x′,t;y1)=0; | (5) |
{∂2tG2−c2ΔG2+ν1∂tG2−ν2∂tΔG2=0,x1,y1>0,x′∈Rn−1,t>0,G2(x1,x′,0;y1)=0,G2t(x1,x′,0;y1)=δ(x1−y1)δ(x′),a1∂x1G2(0,x′,t;y1)+a2G2(0,x′,t;y1)=0. | (6) |
The way of estimating the Green's functions
With the help of the accurate expression of Green's functions for the linear half space problem and the Duhamel's principle, we get the pointwise long time behavior for the nonlinear solution
Theorem 1.1. Let
|∂αxu0,∂αxu1|≤O(1)ε(1+|x|2)−r, r>n2, |α|≤1, |
|∂αxu(x,t)|≤O(1)ε(1+t)−|α|/2(1+t+|x|2)−n2. |
Moreover, we get the following optimal
‖∂αxu(⋅,t)‖Lp(Rn+)≤O(1)ε(1+t)−n2(1−1p)−|α|2, p∈(1,∞]. |
Remark 1. We can develop a similar theorem for the case of higher space dimension with a suitable choice of
Notations. Let
f(ξ,t):=F[f](ξ,t)=∫Rne−iξ⋅xf(x,t)dx,f(x,s):=L[f](x,s)=∫∞0e−stf(x,t)dt. |
The rest of paper is arranged as follows: in Section 2, we study the fundamental solutions for the linear Cauchy problem and give a pointwise description of the fundamental solutions in
The fundamental solutions for the linear damped wave equations are defined by
{∂2tG1−c2ΔG1+ν1∂tG1−ν2∂tΔG1=0G1(x,0)=δ(x),G1t(x,0)=0, | (7) |
{∂2tG2−c2ΔG2+ν1∂tG2−ν2∂tΔG2=0G2(x,0)=0,G2t(x,0)=δ(x). | (8) |
Applying the Fourier transform to (7) and (8) in the space variable
G1(ξ,t)=σ+eσ−t−σ−eσ+tσ+−σ−, G2(ξ,t)=eσ+t−eσ−tσ+−σ−,σ±=−ν1+ν2|ξ|22±12√ν21+(2ν1ν2−4c2)|ξ|2+ν22|ξ|4. |
In [16], authors have studied the pointwise estimates of the fundamental solutions by long wave-short wave decomposition in the Fourier space. Here we will use the local analysis and inverse Fourier transform to get the pointwise structures of the fundamental solutions in the physical variables
f(x,t)=fL(x,t)+fS(x,t),F[fL]=H(1−|ξ|ε0)F[f](ξ,t),F[fS]=(1−H(1−|ξ|ε0))F[f](ξ,t), |
with the parameter
H(x)={1, x>0,0, x<0. |
Long wave component. When
{σ+=−c2|ξ|2ν1+o(|ξ|2),σ−=−ν1+(−ν2+c2ν1)|ξ|2+o(|ξ|2), |
σ+−σ−=ν1+(ν1ν2−2c2)|ξ|2ν1+o(|ξ|2). |
Then
σ+eσ−t=(−c2|ξ|2ν1+o(|ξ|2))e(−ν1+(−ν2+c2ν1)|ξ|2+o(|ξ|2))t=−c2ν1|ξ|2e−ν1t+o(|ξ|2)e−Ct,σ−eσ+t=(−ν1+(−ν2+c2ν1)|ξ|2+o(|ξ|2))e(−c2|ξ|2ν1+o(|ξ|2))t=−ν1e−c2ν1|ξ|2t+O(|ξ|2)e−C|ξ|2t,1σ+−σ−=1ν1+O(|ξ|2). |
So we can approximate the fundamental solutions as follows
σ+eσ−t−σ−eσ+tσ+−σ−=−c2|ξ|2ν21e−ν1t+e−c2ν1|ξ|2t+o(|ξ|2)e−Ct+O(|ξ|2)e−C|ξ|2t,eσ+t−eσ−tσ+−σ−=1ν1e−c2ν1|ξ|2t−1ν1e−ν1t+O(|ξ|2)e−Ct+o(|ξ|2)e−C|ξ|2t. |
Using Lemma 5.1 in Appendix, for
|DαxGL1(x,t)|≤O(1)(e−|x|2C(t+1)(1+t)n+|α|2+e−|x|+tC),|DαxGL2(x,t)|≤O(1)(e−|x|2C(t+1)(1+t)n+|α|2+e−|x|+tC). |
Short wave component. We adopt the local analysis method to give a description about all types of singular functions for the short wave component of the fundamental solutions. When
{σ+=−c2ν2+c2(ν1ν2−c2)ν321|ξ|2+O(|ξ|−4),σ−=−σ+−(ν1+ν2|ξ|2). |
This non-decaying property results in the singularities of the fundamental solution
{σ∗+=−c2ν2+c2(ν1ν2−c2)ν32(11+|ξ|2+1(1+|ξ|2)2)+c2(ν1ν2−c2)ν32O((1+|ξ|2)−3),σ∗−=−σ∗+−(ν1+ν2|ξ|2), |
infξ∈Dε0|σ∗−(ξ)−σ∗+(ξ)|>0,supξ∈Dε0Re(σ∗±(ξ))≤−J0, supξ∈Dε0|ξ|8|σ±(ξ)−σ∗±(ξ)|<∞ as |ξ|→∞. |
Therefore, the approximated analytic spectra
|σ+eσ−t−σ−eσ+tσ+−σ−−σ∗+eσ∗−t−σ∗−eσ∗+tσ∗+−σ∗−, eσ+t−eσ−tσ+−σ−−eσ∗+t−eσ∗−tσ∗+−σ∗−|≤O(1)(1+|ξ|2)4. |
By Lemma 5.4 in the Appendix, we have
‖F−1[σ+eσ−t−σ−eσ+tσ+−σ−−σ∗+eσ∗−t−σ∗−eσ∗+tσ∗+−σ∗−](⋅,t)‖L∞(Rn)=O(1),‖F−1[eσ+t−eσ−tσ+−σ−−eσ∗+t−eσ∗−tσ∗+−σ∗−](⋅,t)‖L∞(Rn)=O(1), |
which asserts that all singularities are contained in
Now we seek out all the singularities. For the short wave part of
σ∗+eσ∗−t−σ∗−eσ∗+tσ∗+−σ∗−=eσ∗+t−σ∗+eσ∗+tσ∗+−σ∗−+σ∗+eσ∗−tσ∗+−σ∗−. |
The first term is
eσ∗+t=e−c2tν2ec2(ν1ν2−c2)tν3211+|ξ|2+c2(ν1ν2−c2)tν321(1+|ξ|2)2+c2(ν1ν2−c2)tν32O(1(1+|ξ|2)3)=e−c2tν2(1+c2(ν1ν2−c2)tν3211+|ξ|2+c2(ν1ν2−c2)tν321(1+|ξ|2)2)+e−c2tν2c2(ν1ν2−c2)tν32O(1(1+|ξ|2)3)=e−c2tν2+c2(ν1ν2−c2)ν32te−c2tν21+|ξ|2+c2(ν1ν2−c2)ν32te−c2tν2(1+|ξ|2)2+te−c2tν2c2(ν1ν2−c2)ν32O(1(1+|ξ|2)3). |
It can be estimated as follows
|F−1[eσ∗+t]−e−c2t/ν2δ(x)−tc2(ν1ν2−c2)ν−32e−c2t/ν2Yn(x)|≤Ce−|x|+tC. |
The second term contains no singularities and we have
σ∗+eσ∗+tσ∗+−σ∗−=−c2ν−22e−c2t/ν21+|ξ|2+e−c2tν2O(1(1+|ξ|2)2), |
so
|F−1[σ∗+eσ∗+tσ∗+−σ∗−]+c2v−22e−c2t/v2Yn(x)|≤Ce−|x|+tC. |
For the third term, the function
|σ∗+eσ∗−tσ∗+−σ∗−|≤K0e−|ξ|2t/C1−J∗0t1+|ξ|2, |
∫Im(ξk)=δ1≤k≤n|σ∗+eσ∗−tσ∗+−σ∗−|dξ≤C∫Rne−|ξ|2t/C−J∗0t(1+|ξ|)2dξ=CΓ(n)∫∞0e−r2t/C−J∗0t(1+r)2rn−1dr≤Ce−t/CLn(t), | (9) |
where
Ln(t)≡{1,n=1,log(t),n=2,t−n−22,n≥3. |
We denote
j1(x,t):=F−1[σ∗+eσ∗−tσ∗+−σ∗−], |
following the way of proof for Lemma 5.4, we get
|j1(x,t)|≤Ce−(|x|+t)/CLn(t) |
from (9). So the following estimate for
|GS1(x,t)−j1(x,t)−e−c2t/ν2δn(x)−(tc2ν−32(ν1ν2−c2)+c2ν−22)e−c2t/ν2Yn(x)|≤e−|x|+tC. |
For the short wave part of
eσ∗+t−eσ∗−tσ∗+−σ∗−=eσ∗+tσ∗+−σ∗−−eσ∗−tσ∗+−σ∗−. |
The first term is
eσ∗+tσ∗+−σ∗−=ν−12e−c2t/ν21+|ξ|2+e−c2tν2O(1(1+|ξ|2)2), |
and we have
|F−1[eσ∗+tσ∗+−σ∗−]−ν−12e−c2t/ν2Yn(x)|≤Ce−|x|+tC. |
The second term contains no singularities. If denoting
j2(x,t)≡−F−1(eσ∗−tσ∗+−σ∗−), |
then there exists
|j2(x,t)|≤Ce−(|x|+t)/CLn(t), |
and we have the following estimate for
|GS2(x,t)−j2(x,t)−ν−12e−c2t/ν2Yn(x)|≤Ce−|x|+tC. |
Hence the short wave components have the following estimates in the finite Mach number region
|GS1(x,t)−j1(x,t)−e−c2tν2δn(x)−(tc2(ν1ν2−c2)ν32+c2ν22)e−c2tν2Yn(x)|≤Ce−|x|+tC.|GS2(x,t)−j2(x,t)−ν−12e−c2tν2Yn(x)|≤Ce−|x|+tC. |
Outside the finite Mach number region
We choose the weighted function
wt=−aMw,∇w=xM|x|w,Δw=wM2. |
Consider the linear damped wave equation outside the finite Mach number region:
{∂2tui−c2Δui+ν1∂tui−ν2∂tΔui=0,|x|≥3(t+1),ui|t=0=0,uit|t=0=0,ui||x|=3(t+1)=Gi||x|=3(t+1). | (10) |
Denote the outside finite Mach number region
c2∫∂Dtw∂tui∇ui⋅d→Sx+ν2∫∂Dtw∂tui∂t∇ui⋅d→Sx=12ddt∫Dtw((∂tui)2+c2|∇ui|2)dx+∫Dt(ν1w−12wt−12ν2Δw)(∂tui)2dx+c2∫Dt∂tui∇w⋅∇uidx+ac22M∫Dtw|∇ui|2dx+ν2∫Dtw|∂t∇ui|2dx=12ddt∫Dtw((∂tui)2+c2|∇ui|2)dx+∫Dt(ν1+a2M−ν22M2)w(∂tui)2dx+c2∫Dtw∂tuixM|x|⋅∇uidx+ac22M∫Dtw|∇ui|2dx+ν2∫Dtw|∂t∇ui|2dx≥12ddt∫Dtw((∂tui)2+c2|∇ui|2)dx+∫Dtw(ac24M|∇ui|2+(ν12+a2M−ν22M2)(∂tui)2+ν2|∂t∇ui|2)dx. |
On the boundary
|∂tui|,|∇ui|,|∂t∇ui|≤Ce−Ct, x∈∂Dt. |
So
ddt∫Dtw((∂tui)2+c2|∇ui|2)dx+2δ0∫Dtw((∂tui)2+c2|∇ui|2)dx≤Ce−Ct, | (11) |
One can also get similar estimates for any higher order derivatives
l∑|α|=1(ddt∫Rnw((∂t∂αxui)2+c2|∇∂αxui|2)dx) +δ|α|∫Rnw((∂t∂αxui)2+c2|∇∂αxui|2)dx)≤Ce−Ct. | (12) |
Integrating (11) and (12) over
sup(x,t)∈Dt((∂t∂αxui)2+c2|∇∂αxui|2)≤Ce−(|x|−at)/C≤Ce−(|x|+t)/C, for |α|<l−n2, |
since
|DαxGi(x,t)|≤Ce−(|x|+t)/C, for |α|<l−n2. |
To summarize, we have the following pointwise estimates for the fundamental solutions:
Lemma 2.1. The fundamental solutions have the following estimates for all
|Dαx(G1(x,t)−j1(x,t)−e−c2t/ν2δn(x)−(tc2ν−32(ν1ν2−c2)+c2ν−22)e−c2t/ν2Yn(x))|≤O(1)(e−|x|2C(t+1)(t+1)n+|α|2+e−(|x|+t)/C),|Dαx(G2(x,t)−j2(x,t)−ν−12e−c2t/ν2Yn(x))|≤O(1)(e−|x|2C(t+1)(t+1)n+|α|2+e−(|x|+t)/C). |
Here
|j1(x,t),j2(x,t)|≤O(1)Ln(t)e−(|x|+t)/C,L2(t)=log(t), Ln(t)=t−n−22 for n≥3,Y2(x)=O(1)12πBesselK0(|x|), Yn(x)=O(1)e−|x||x|n−2 for n≥3. |
Applying Laplace transform in
G1(ξ,s)=s+ν1+ν2|ξ|2s2+ν1s+(c2+ν2s)|ξ|2, G2(ξ,s)=1s2+ν1s+(c2+ν2s)|ξ|2. |
Now we give a lemma:
Lemma 2.2.
12π∫Reiξ1x1s2+ν1s+ν2s|ξ|2+c2|ξ|2dξ1=1ν2s+c2e−λ|x1|2λ, |
where
Proof. We prove it by using the contour integral and the residue theorem. Note that
12π∫Reiξ1x1s2+ν1s+ν2s|ξ|2+c2|ξ|2dξ1=12π1ν2s+c2∫Reiξ1x1ξ21+|ξ′|2+s2+ν1sν2s+c2dξ1=12π1ν2s+c2∫Reiξ1x1(ξ1−λi)(ξ1+λi)dξ1. |
Define a closed path
If
12π1ν2s+c2∫Reiξ1x1(ξ1−λi)(ξ1+λi)dξ1=12π1ν2s+c22πiRes(eiξ1x1(ξ1−λi)(ξ1+λi)|ξ1=λi)=e−λx12(ν2s+c2)λ. |
The computation for the case
12π1ν2s+c2∫Reiξ1x1(ξ1−λi)(ξ1+λi)dξ1=−12π1ν2s+c22πiRes(eiξ1x1(ξ1−λi)(ξ1+λi)|ξ1=−λi)=eλx12(ν2s+c2)λ. |
Hence we prove this lemma.
With the help of Lemma 2.2, we get the expression of fundamental solutions
G1(x1,ξ′,s)=1ν2s+c2(ν2δ(x1)+c2(s+ν1)ν2s+c2e−λ|x1|2λ),G2(x1,ξ′,s)=e−λ|x1|2λ(ν2s+c2). |
In particular, when
G1(−ˉx1,ξ′,s)=c2(s+ν1)(ν2s+c2)2e−λˉx12λ, G2(−ˉx1,ξ′,s)=e−λˉx12λ(ν2s+c2). |
In this section, we will give the pointwise estimates of the Green's functions for the initial boundary value problem. Firstly, we compute the transformed Green's functions in the partial-Fourier and Laplace transformed space. Then by comparing the symbols of the fundamental solutions and the Green's functions in this transformed space, we get the simplified expressions of Green's functions for the initial-boundary value problem. With the help of the pointwise estimates of the fundamental solutions and boundary operator, we finally get the sharp estimates of Green functions for the half space linear problem.
Before computing, we make the initial value zero by considering the error function
{∂2tRi−c2ΔRi+ν1∂tRi−ν2∂tΔRi=0,x∈Rn+,t>0,Ri|t=0=0,Rit|t=0=0,(a1∂x1+a2)Ri(0,x′,t;y1)=−(a1∂x1+a2)Gi(x1−y1,x′,t)|x1=0. |
Taking Fourier transform only with respect to the tangential spatial variable
{(s2+ν1s)Ri−(c2+ν2s)Rix1x1+(c2+ν2s)|ξ′|2Ri=0,(a1∂x1+a2)Ri(0,ξ′,s;y1)=(a1∂y1−a2)Gi(−y1,ξ′,s)=−(a1λ+a2)Gi(−y1,ξ′,s). |
Solving it and dropping out the divergent mode as
Ri(x1,ξ′,s;y1)=−a1λ+a2a2−a1λe−λx1Gi(−y1,ξ′,s)=−a1λ+a2a2−a1λGi(x1+y1,ξ′,s), |
where
Therefore the transformed Green's functions
Gi(x1,ξ′,s;y1)=Gi(x1−y1,ξ′,s)−a1λ+a2a2−a1λGi(x1+y1,ξ′,s)=Gi(x1−y1,ξ′,s)+Gi(x1+y1,ξ′,s)−2a2a2−a1λGi(x1+y1,ξ′,s), |
which reveal the connection between fundamental solutions and the Green's functions.
Hence,
Gi(x1,x′,t;y1)=Gi(x1−y1,x′,t)+Gi(x1+y1,x′,t)−F−1ξ′→x′L−1s→t[2a2a2−a1λ]∗x′,tGi(x1+y1,x′,t). |
Now we estimate the boundary operator
Instead of inverting the boundary symbol, we follow the differential equation method. Notice that
F−1ξ′→x′L−1s→t[2a2a2−a1λGi(x1+y1,ξ′,s)]=2a2a1∂x1+a2Gi(x1+y1,x′,t), |
setting
g(x1,x′,t)≡2a2a1∂x1+a2Gi(x1,x′,t), |
then the function
(a2+a1∂x1)g=2a2Gi(x1,x′,t). |
Solving this ODE gives
g(x1,x′,t)=2γ∫∞x1e−γ(z−x1)Gi(z,x′,t)dz=2γ∫∞0e−γzGi(x1+z,x′,t)dz. | (13) |
Summarizing previous results we obtain
Lemma 3.1. The Green's functions
Gi(x1,x′,t;y1)=GLi(x1,x′,t;y1)+GSi(x1,x′,t;y1). |
Meanwhile, the following estimates hold:
|DαxGLi(x1,x′,t;y1)|≤O(1)(e−(x1−y1)2+(x′−y′)2C(t+1)(t+1)n+|α|2+e−(x1+y1)2+(x′−y′)2C(t+1)(t+1)n+|α|2),|α|≥0; |
|GS1(x1,x′,t;y1)|≤O(1)(j1(x1−y1,x′,t)+j1(x1+y1,x′,t)+e−c2tν2(δn(x1−y1,x′)+δn(x1+y1,x′))+e−c2tν2(tc2ν−32(ν1ν2−c2)+c2ν−22)(Yn(x1−y1,x′)+Yn(x1+y1,x′))) |
and
|GS2(x1,x′,t;y1)|≤O(1)(j1(x1−y1,x′,t)+j2(x1+y1,x′,t)+ν−12e−c2tν2(Yn(x1−y1,x′)+Yn(x1+y1,x′))). |
Proof. Note that
Gi(x1,x′,t;y1)=Gi(x1−y1,x′,t)+Gi(x1+y1,x′,t)−g(x1+y1,x′,t), |
based on the long-wave short-wave decomposition of the fundamental solutions
Gi(x,t)=GLi(x,t)+GSi(x,t), |
we can write
GLi(x1,x′,t;y1)=O(1)(GLi(x1−y1,x′,t)+GLi(x1+y1,x′,t)),GSi(x1,x′,t;y1)=O(1)(GSi(x1−y1,x′,t)+GSi(x1+y1,x′,t)), |
and get the estimates directly from Lemma 2.1 and (13).
The study of boundary operator in the last section suggests that we can only consider the case
Now we give the pointwise long time behavior of the solution for the nonlinear problem and prove the Theorem 1.1. The Green's functions
∂αxu(x,t)=∂αx∫Rn+(G1(x1,x′−y′,t;y1)u0(y)+G2(x1,x′−y′,t;y1)u1(y))dy+∂αx∫t0∫Rn+G2(x1,x′−y′,t−τ;y1)f(u)(y,τ)dydτ≡∂αxI(x,t)+∂αxN(x,t). | (14) |
The initial part
∂αxI(x,t)=∂αxIL(x,t)+∂αxIS(x,t), |
where
∂αxIL(x,t)=∂αx∫Rn+(GL1(x1,x′−y′,t;y1)u0(y)+GL2(x1,x′−y′,t;y1)u1(y))dy∂αxIS(x,t)=∂αx∫Rn+(GS1(x1,x′−y′,t;y1)u0(y)+GS2(x1,x′−y′,t;y1)u1(y))dy. |
By lemma 5.2, we have the following estimates in the finite Mach number region
|IL(x,t)|≤O(1)ε∫Rn+e−(x−y)2C(t+1)(t+1)n2(1+|y|2)−rdy≤O(1)ε(e−x2C(t+1)(t+1)n2+(1+t+|x|2)−n2), | (15) |
\begin{align} \label{{4.3b}} \begin{aligned} &|\mathcal{I}^S({\bf{x}},t)|\\ \le& O(1)\varepsilon e^{-\frac{(|{\bf{x}}|+t)}{C}}\Bigg|\int_{\mathbb{R}^n}\Big(L_n(t) +\delta_n(\bf{x-y})\\ & \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ +\left[\frac{tc^2}{\nu_2^{3}}(\nu_1\nu_2-c^2)+\frac{c^2}{\nu_2^{2}}\right]Y_n(\bf{x-y})\Big)\left(1+|{\bf{y}}|^2\right)^{-r}d{\bf{y}} \Bigg|\\ &+O(1)\varepsilon e^{-\frac{(|{\bf{x}}|+t)}{C}}\left|\int_{\mathbb{R}^n}\left(L_n(t)+\nu_2^{-1}Y_n(\bf{x-y})\right)\left(1+|{\bf{y}}|^2\right)^{-r}d{\bf{y}}\right|\\ \le &O(1)\varepsilon\left(\frac{e^{-\frac{{\bf{x}}^2}{C(t+1)}}}{(t+1)^{\frac{n}{2}}}+\left(1+t+\left|{\bf{x}}\right|^2\right)^{-\frac{n}{2}}\right). \end{aligned} \end{align} | (16) |
Hence we combine (15) and (16) to get the estimate of the first part in (14) when
\begin{align} \begin{aligned} \label{{4.3}} \left|\mathcal{I}({\bf{x}},t)\right| \le O(1)\varepsilon\left(\frac{e^{-\frac{{\bf{x}}^2}{C(t+1)}}}{(t+1)^{\frac{n}{2}}}+\left(1+t+\left|{\bf{x}}\right|^2\right)^{-\frac{n}{2}}\right). \end{aligned} \end{align} | (17) |
Similarly, when
\begin{align*} \begin{aligned} &\left|\partial_{{\bf{x}}}^\alpha\mathcal{I}({\bf{x}},t)\right| = \left|\partial_{{\bf{x}}}^\alpha\mathcal{I}^L({\bf{x}},t)+\partial_{{\bf{x}}}^\alpha\mathcal{I}^S({\bf{x}},t)\right|\\ \le&O(1)\varepsilon\int_{ \mathbb{R}^n_+}\left(\frac{e^{-\frac{(x_1-y_1)^2+({\bf{x'-y'}})^2}{C(t+1)}}}{(t+1)^{\frac{n}{2}+\frac{1}{2}}}+\frac{e^{-\frac{(x_1+y_1)^2+({\bf{x'-y'}})^2}{C(t+1)}}}{(t+1)^{\frac{n}{2}+\frac{1}{2}}}\right)\left(1+|{\bf{y}}|^2\right)^{-r}d{\bf{y}}\\ &+1_{\{\partial^\alpha_{{\bf{x}}} = \partial_{x_1}\}}O(1)\varepsilon e^{-\frac{(|{\bf{x}}|+t)}{C}}\Bigg|\int_{\mathbb{R}^{n-1}}L_n(t)+\delta_n(x_1-y_1,{\bf{x'-y'}},t)\\ & \ \ \ \ \ \ +\delta_n(x_1+y_1,{\bf{x'-y'}},t)+\left(tc^2\nu_2^{-3}(\nu_1\nu_2-c^2)+c^2\nu_2^{-2}\right)\\ &\ \ \ \ \ \left(Y_n(x_1-y_1,{\bf{x'-y'}})+Y_n(x_1+y_1,{\bf{x'-y'}})\right)\left(1+|{\bf{y}}|^2\right)^{-r}d\bf{y'}|_{y_1 = 0}\Bigg|\\ &+O(1)\varepsilon e^{-\frac{(|{\bf{x}}|+t)}{C}}\Bigg|\int_{\mathbb{R}_+^n}L_n(t)+\delta_n(x_1-y_1,{\bf{x'-y'}},t)+\delta_n(x_1+y_1,{\bf{x'-y'}},t)\\ & \ \ \ \ \ +\left(tc^2\nu_2^{-3}(\nu_1\nu_2-c^2)+c^2\nu_2^{-2}\right)\\ & \ \ \ \ \ \ \left(Y_n(x_1-y_1,{\bf{x'-y'}})+Y_n(x_1+y_1,{\bf{x'-y'}})\right)\left(1+|{\bf{y}}|^2\right)^{-r}d{\bf{y}}\Bigg|\\ &+1_{\{\partial^\alpha_{{\bf{x}}} = \partial_{x_1}\}}O(1)\varepsilon e^{-\frac{(|{\bf{x}}|+t)}{C}}\Bigg|\int_{\mathbb{R}^{n-1}}(L_n(t)+\nu_1^{-1}Y_n(x_1-y_1,{\bf{x'-y'}})\\ & \ \ \ \ +\nu_1^{-1}Y_n(x_1+y_1,{\bf{x'-y'}}))\left(1+|{\bf{y}}|^2\right)^{-r}d\bf{y'}|_{y_1 = 0}\Bigg|\\ &+O(1)\varepsilon e^{-\frac{(|{\bf{x}}|+t)}{C}}\Bigg|\int_{\mathbb{R}^{n-1}}(L_n(t)+\nu_1^{-1}Y_n(x_1-y_1,{\bf{x'-y'}})\\ & \ \ \ \ +\nu_1^{-1}Y_n(x_1+y_1,{\bf{x'-y'}}))\left(1+|{\bf{y}}|^2\right)^{-r}d{\bf{y}}\Bigg|\\ \le&O(1)\varepsilon(1+t)^{-\frac{|\alpha|}{2}}\left(\frac{e^{-\frac{{\bf{x}}^2}{2C(t+1)}}}{(t+1)^{\frac{n}{2}}}+\left(1+t+\left|{\bf{x}}\right|^2\right)^{-r}\right)+O(1)\varepsilon e^{-(|{\bf{x}}|+t)/C}. \end{aligned} \end{align*} |
where
\begin{align*} \begin{aligned} 1_{\{\partial^\alpha_{{\bf{x}}} = \partial_{x_1}\}} = \left\{\begin{array}{lll} 1,\ if\ \partial^\alpha_{{\bf{x}}} = \partial_{x_1},\\ 0,\ otherwise.\end{array}\right. \end{aligned} \end{align*} |
Here we use the integration by parts to estimate the short wave component part. Outside the finite Mach number region, we have
\begin{align} \label{{4.5}} \begin{aligned} &\left|\partial_{{\bf{x}}}^\alpha\mathcal{I}({\bf{x}},t)\right|\le O(1)\varepsilon e^{-\nu_1t}\int_{ \mathbb{R}_+^n}e^{-|\bf{x-y}|}(1+{\bf{y}}^2)^{-r}d{\bf{y}}\\ \le& O(1)\varepsilon e^{-\nu_1t}(1+|{\bf{x}}|^2)^{-r},\quad|\alpha|\le1. \end{aligned} \end{align} | (18) |
Based on the estimates of (17)-(18), the ansatz is posed for the solution as follows:
\begin{align*} \begin{aligned} \left|\partial_{{\bf{x}}}^\alpha u({\bf{x}},t)\right| \le O(1)\varepsilon (1+t)^{-\frac{|\alpha|}{2}}(1+t+|{\bf{x}}|^2)^{-\frac{n}{2}}, \quad |\alpha|\le1. \end{aligned} \end{align*} |
Straightforward computations show that
\begin{eqnarray*} \left|f(u)({\bf{x}},t)\right|\le O(1)\varepsilon^k(1+t+|{\bf{x}}|^2)^{-\frac{nk}{2}}. \end{eqnarray*} |
Now we justify the ansatz for the nonlinear term. For
\begin{align*} \begin{aligned} \left|\mathcal{N}({\bf{x}},t)\right| = &\left|\int_{0}^{t}\int_{ \mathbb{R}^n_+}\mathbb{G}_2(x_1,{\bf{x'-y'}},t-\tau;y_1)f(u)({\bf{y}},\tau)d{\bf{y}}d\tau\right|\\ \le&\left|\int_{0}^{t}\int_{\mathbb{R}^n_+}\mathbb{G}_2^L(x_1,{\bf{x'-y'}},t-\tau;y_1)f(u)({\bf{y}},\tau)d{\bf{y}}d\tau\right|\\&+\left|\int_{0}^{t}\int_{\mathbb{R}^n_+}\mathbb{G}_2^S(x_1,{\bf{x'-y'}},t-\tau;y_1)f(u)({\bf{y}},\tau)d{\bf{y}}d\tau\right|\\ = &\mathcal{N}_1+\mathcal{N}_2. \end{aligned} \end{align*} |
Using Lemma 5.3, one gets
\begin{align*} \begin{aligned} \mathcal{N}_1&\le O(1)\varepsilon^k\Bigg|\int^t_0\int^\infty_0\int_{\mathbb{R}^{n-1}}\left(\frac{e^{-\frac{(x_1-y_1)^2+({\bf{x'-y'}})^2}{C(t-\tau+1)}}}{(t-\tau+1)^{\frac{n}{2}}}+\frac{e^{-\frac{(x_1+y_1)^2+({\bf{x'-y'}})^2}{C(t-\tau+1)}}}{(t-\tau+1)^{\frac{n}{2}}}\right)\\ &\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ (1+\tau+|{\bf{y}}|^2)^{-\frac{nk}{2}}d{\bf{y}}'dy_1d\tau\Bigg|\\ &\le O(1)\varepsilon^k\left|\int^t_0\int_{ \mathbb{R}^n}\frac{e^{-\frac{(x_1-y_1)^2+({\bf{x'-y'}})^2}{C(t-\tau+1)}}}{(t-\tau+1)^{\frac{n}{2}}}(1+\tau+|{\bf{y}}|^2)^{-\frac{nk}{2}}d{\bf{y}}d\tau\right|\\ &\le O(1)\varepsilon^k(1+t+|{\bf{x}}|^2)^{-\frac{n}{2}}, \end{aligned} \end{align*} |
\begin{align*} \begin{aligned} \mathcal{N}_2&\le O(1)\varepsilon^k\Bigg|\int^t_0\int_{ \mathbb{R}^n_+}e^{-\frac{c^2(t-\tau)}{\nu_2}}\left(L_n(t-\tau)+\nu_2^{-1}Y_n(x_1,{\bf{x'-y'}};y_1)\right)\\ &\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ (1+\tau+|{\bf{y}}|^2)^{-\frac{nk}{2}}d{\bf{y}}d\tau\Bigg|\\ &\le O(1)\varepsilon^k(1+t+|{\bf{x}}|^2)^{-\frac{n}{2}}. \end{aligned} \end{align*} |
Now we compute the estimate of
\begin{align*} \begin{aligned} \left|\partial_{{\bf{x}}}^\alpha\mathcal{N}({\bf{x}},t)\right| = &\left|\partial_{{\bf{x}}}^\alpha\int_{0}^{t}\int_{ \mathbb{R}^n_+}\mathbb{G}_2(x_1,{\bf{x'-y'}},t-\tau;y_1)f(u)({\bf{y}},\tau)d{\bf{y}}d\tau\right|\\ \le&\left|\int_{0}^{t}\int_{\mathbb{R}^n_+}\partial_{{\bf{x}}}^\alpha\mathbb{G}_2^L(x_1,{\bf{x'-y'}},t-\tau;y_1)f(u)({\bf{y}},\tau)d{\bf{y}}d\tau\right|\\ &+\left|\int_{0}^{t}\int_{\mathbb{R}^n_+}\partial_{{\bf{x}}}^\alpha\mathbb{G}_2^S(x_1,{\bf{x'-y'}},t-\tau;y_1)f(u)({\bf{y}},\tau)d{\bf{y}}d\tau\right|\\ = &\partial_{{\bf{x}}}^\alpha\mathcal{N}_1+\partial_{{\bf{x}}}^\alpha\mathcal{N}_2. \end{aligned} \end{align*} |
Similarly we have
\begin{align*} \begin{aligned} \partial_{{\bf{x}}}^\alpha\mathcal{N}_1 = &\Bigg|O(1)\varepsilon^k\int^t_0\int^\infty_0\int_{\mathbb{R}^{n-1}}\left(\frac{e^{-\frac{(x_1-y_1)^2+({\bf{x'-y'}})^2}{C(t-\tau+1)}}}{(t-\tau+1)^{\frac{n}{2}+\frac{|\alpha|}{2}}}+\frac{e^{-\frac{(x_1+y_1)^2+({\bf{x'-y'}})^2}{C(t-\tau+1)}}}{(t-\tau+1)^{\frac{n}{2}+\frac{|\alpha|}{2}}}\right)\\ &\ \ \ \ \ (1+\tau+|{\bf{y}}|^2)^{-\frac{nk}{2}}d{\bf{y}}'dy_1d\tau\Bigg|\\ \le&\left|O(1)\varepsilon^k\int^t_0\int_{ \mathbb{R}^n}\frac{e^{-\frac{(x_1-y_1)^2+({\bf{x'-y'}})^2}{C(t-\tau+1)}}}{(t-\tau+1)^{\frac{n}{2}+\frac{|\alpha|}{2}}}(1+\tau+|{\bf{y}}|^2)^{-\frac{nk}{2}}d{\bf{y}}d\tau\right|\\ \le&O(1)\varepsilon^k(1+t)^{-\frac{|\alpha|}{2}}(1+t+|{\bf{x}}|^2)^{-\frac{n}{2}}, \end{aligned} \end{align*} |
\begin{align} \label{{10.0}} \begin{aligned} \partial_{{\bf{x}}}^\alpha\mathcal{N}_2 = &\left|\int_{0}^{t}\int_{\mathbb{R}^n_+}\partial_{{\bf{x}}}^\alpha\mathbb{G}_2^S(x_1,{\bf{x'-y'}},t-\tau;y_1)f(u)({\bf{y}},\tau)d{\bf{y}}d\tau\right|\\ = &1_{\{\partial_{{\bf{x}}}^\alpha = \partial_{x_1}\}}\left|\int_{0}^{t}\int_{\mathbb{R}^{n-1}}\mathbb{G}_2^S(x_1,{\bf{x'-y'}},t-\tau;y_1) f(u)({\bf{y}},\tau)d{\bf{y}}'|_{y_1 = 0}d\tau\right|\\ &+\left|\int_{0}^{t}\int_{\mathbb{R}^n_+}\mathbb{G}_2^S(x_1,{\bf{x'-y'}},t-\tau;y_1)\partial_{{\bf{y}}}^\alpha f(u)({\bf{y}},\tau)d{\bf{y}}d\tau\right|. \end{aligned} \end{align} | (19) |
The boundary term in (19) has the following estimates:
\begin{align*} \begin{aligned} &\left|\int_{0}^{t}\int_{\mathbb{R}^{n-1}}\mathbb{G}_2^S(x_1,{\bf{x'-y'}},t-\tau;y_1)f(u)({\bf{y}},\tau)d{\bf{y}}'|_{y_1 = 0}d\tau\right|\\ \le& \left|\left(\int_{0}^{t/2}+\int_{t/2}^t\right)\int_{\mathbb{R}^{n-1}}\mathbb{G}_2^S(x_1,{\bf{x'-y'}},t-\tau;y_1)f(u)({\bf{y}},\tau)d{\bf{y}}'|_{y_1 = 0}d\tau\right|\\ \le&O(1)\varepsilon^k(1+t)^{-\frac{|\alpha|}{2}}(1+t+|{\bf{x}}|^2)^{-\frac{n}{2}}. \end{aligned} \end{align*} |
The second term in (19) satisfies
\begin{align*} \begin{aligned} &\left|\int_{0}^{t}\int_{\mathbb{R}^n_+}\mathbb{G}_2^S(x_1,{\bf{x'-y'}},t-\tau;y_1)\partial_{{\bf{y}}}^\alpha f(u)({\bf{y}},\tau)d{\bf{y}}d\tau\right|\\ \le&\Bigg|O(1)\varepsilon^k\int^t_0\int_{ \mathbb{R}^n_+}e^{-\frac{c^2(t-\tau)}{\nu_2}}\left(L_n(t-\tau)+\nu_2^{-1}Y_n(x_1,{\bf{x'-y'}};y_1)\right)\\ \le& (1+\tau)^{-\frac{|\alpha|}{2}}(1+\tau+|{\bf{y}}|^2)^{-\frac{nk}{2}}d{\bf{y}}d\tau\Bigg|\\ \le& O(1)\varepsilon^k(1+t)^{-\frac{|\alpha|}{2}}(1+t+|{\bf{x}}|^2)^{-\frac{n}{2}}. \end{aligned} \end{align*} |
Therefore one has the following estimate for the nonlinear term
\begin{align*} \begin{aligned} \left|\partial_{{\bf{x}}}^\alpha\mathcal{N}\right|\le O(1)\varepsilon^k(1+t)^{-\frac{|\alpha|}{2}}(1+t+|{\bf{x}}|^2)^{-\frac{n}{2}}, \quad|\alpha|\le1. \end{aligned} \end{align*} |
Outside the finite Mach number region,
\begin{align*} \begin{aligned} \left|\partial_{{\bf{x}}}^\alpha\mathcal{N}\right|\le& O(1)\varepsilon^k\left|\int^t_0\int_{ \mathbb{R}^n_+}e^{-\nu_1(t-\tau)}e^{-|\bf{x-y}|}(1+\tau+|{\bf{y}}^2|)^{-\frac{nk}{2}}d{\bf{y}}d\tau\right|\\ \le&O(1)\varepsilon^k(1+t+|{\bf{x}}|^2)^{-\frac{nk}{2}}, \quad |\alpha|\le1. \end{aligned} \end{align*} |
Thus, we verify the ansatz and finish the proof of pointwise estimates of the solution.
The
\begin{align*} \begin{aligned} &\left(\int_{\mathbb{R}^{n}_+}(1+t+|{\bf{x}}|^2)^{-\frac{n}{2}p} d{\bf{x}}\right)^{\frac{1}{p}} = \left(\int_{\mathbb{R}^{n}_+}(1+t)^{-\frac{n}{2}p}\left(1+\frac{|{\bf{x}}|^2}{1+t}\right)^{-\frac{n}{2}p}d{\bf{x}}\right)^{\frac{1}{p}}\\ = &(1+t)^{-\frac{n}{2}}(1+t)^{\frac{n}{2p}} = (1+t)^{-\frac{n}{2}(1-\frac{1}{p})}. \end{aligned} \end{align*} |
Hence we finish the proof of Theorem 1.1.
Lemma 5.1. [10] In the finite Mach number region
\begin{align*} \begin{aligned} \left|\frac{1}{(2\pi)^n}\int_{|{\boldsymbol{\xi}}|\le\varepsilon_0}(i{\boldsymbol{\xi}})^\alpha e^{i {\boldsymbol{\xi}}\cdot {\bf{x}}}e^{-\frac{1}{\kappa}|{\boldsymbol{\xi}}|^2t}d{\boldsymbol{\xi}}\right| \le O(1)\frac{e^{-\frac{|{\bf{x}}|^2}{C(t+1)}}}{(1+t)^{\frac{n+|\alpha|}{2}}}+O(1)e^{-\frac{|{\bf{x}}|+t}{C}}, |\alpha|\ge 0. \end{aligned} \end{align*} |
Lemma 5.2. [9] We have the follow estimate for
\begin{align*} \begin{aligned} &\int_{\mathbb R^n}\frac{e^{-\frac{({\bf{x}}-{\bf{y}})^2}{C(t+1)}}}{(1+t)^{\frac{n}{2}+\frac{|\alpha|}{2}}}(1+|{\bf{y}}|^2)^{-r}d{\bf{y}}\le O(1)(1+t)^{-\frac{|\alpha|}{2}}\left(\frac{e^{-\frac{{\bf{x}}^2}{2C(t+1)}}}{(t+1)^{\frac{n}{2}}}+(1+t+|{\bf{x}}|^2)^{-r}\right). \end{aligned} \end{align*} |
Lemma 5.3. [9] For
\begin{align*} \begin{aligned} &\int_{0}^{t}\int_{ \mathbb{R}^n}e^{-\frac{\nu(t-\tau)}{2}}Y_n(\bf{x-y})(1+\tau)^{-\frac{|\alpha|}{2}}(1+\tau+|{\bf{y}}|^2)^\frac{nk}{2}d{\bf{y}}d\tau\\ \le& O(1)(1+t)^{-\frac{|\alpha|}{2}}(1+t+|{\bf{x}}|)^{-nk/2}, \end{aligned} \end{align*} |
\begin{align*} \begin{aligned} \int_{0}^{t}\int_{\mathbb{R}^n}\frac{e^{-\frac{({\bf{x}}-{\bf{y}})^2}{C(t-\tau+1)}}}{(1+t)^{\frac{n}{2}+\frac{|\alpha|}{2}}}(1+\tau+|{\bf{y}}|^2)^{-\frac{nk}{2}}d{\bf{y}}d\tau\le O(1)(1+t)^{-\frac{|\alpha|}{2}}(1+t+|{\bf{x}}|^2)^{-\frac{n}{2}}. \end{aligned} \end{align*} |
Lemma 5.4. [7] Suppose a function
\begin{align*} \begin{aligned} |\mathcal{F}[f]({\boldsymbol{\xi}})|\le \frac{E}{(1+|{\boldsymbol{\xi}}|)^{n+1}}, \ \ \mathit{\mbox{for}} \ \ |Im(\xi_i)|\le\delta, \ \ \mathit{\mbox{and}}\ \ i = 1,2,\cdots, n. \end{aligned} \end{align*} |
Then, the function
\begin{eqnarray*} |f({\bf{x}})|\le Ee^{-\delta |{\bf{x}}|/C}, \end{eqnarray*} |
for any positive constant
The authors would like to thank the referees very much for their valuable comments and suggestions which improve the presentation of papersignicantly.
[1] |
C. Phua, D. Alahakoon, V. Lee, Minority report in fraud detection: classification of skewed data, ACM SIGKDD Explor. Newsl., 6 (2004), 50–59. https://doi.org/10.1145/1007730.1007738 doi: 10.1145/1007730.1007738
![]() |
[2] |
B. Krawczyk, M. Galar, L. Jelen, F. Herrera, Evolutionary undersampling boosting for imbalanced classification of breast cancer malignancy, Appl. Soft Comput., 38 (2016), 714–726. https://doi.org/10.1016/j.asoc.2015.08.060 doi: 10.1016/j.asoc.2015.08.060
![]() |
[3] |
J. Alqatawna, H. Faris, K. Jaradat, M. Al-Zewairi, O. Adwan, Improving knowledge based spam detection methods: The effect of malicious related features in imbalance data distribution, Int. J. Commun. Network Syst. Sci., 8 (2015), 118–129. https://doi.org/10.4236/ijcns.2015.85014 doi: 10.4236/ijcns.2015.85014
![]() |
[4] |
N. Japkowicz, S. Stephen, The class imbalance problem: A systematic study, Intell. Data Anal., 6 (2002), 429–449. https://doi.org/10.3233/IDA-2002-6504 doi: 10.3233/IDA-2002-6504
![]() |
[5] | X. Fan, H. Yu, GAMC: An oversampling method based on genetic algorithm and monte carlo method to solve the class imbalance issue in industry, in 2022 International Conference on Industrial IoT, Big Data and Supply Chain (IIoTBDSC), (2022), 127–132. https://doi.org/10.1109/IIoTBDSC57192.2022.00033 |
[6] | F. Zhang, G. Liu, Z. Li, C. Yan, C. Jang, GMM-based undersampling and its application for credit card fraud detection, in 2019 International Joint Conference on Neural Networks (IJCNN), (2019), 1–8. https://doi.org/10.1109/IJCNN.2019.8852415 |
[7] |
Y. Yan, Y. Zhu, R. Liu, Y. Zhang, Y. Zhang, L. Zhang, Spatial distribution-based imbalanced undersampling, IEEE Trans. Knowl. Data Eng., 35 (2023), 6376–6391. https://doi.org/10.1109/TKDE.2022.3161537 doi: 10.1109/TKDE.2022.3161537
![]() |
[8] |
H. Zhu, M. Zhou, G. Liu, Y. Xie, S. Liu, C. Guo, NUS: Noisy-sample-removed undersampling scheme for imbalanced classification and application to credit card fraud detection, IEEE Trans. Comput. Soc. Syst., (2023), 1–12. https://doi.org/10.1109/TCSS.2023.3243925 doi: 10.1109/TCSS.2023.3243925
![]() |
[9] |
N. V. Chawla, K. W. Bowyer, L. O. Hall, W. P. Kegelmeyer, SMOTE: Synthetic minority over-sampling technique, J. Artif. Intell. Res., 16 (2002), 321–357. https://doi.org/10.1613/jair.953 doi: 10.1613/jair.953
![]() |
[10] |
A. Fernández, S. Garcia, F. Herrera, N. V. Chawla, SMOTE for learning from imbalanced data: progress and challenges, marking the 15-year anniversary, J. Artif. Intell. Res., 61 (2018), 863–905. https://doi.org/10.1613/jair.1.11192 doi: 10.1613/jair.1.11192
![]() |
[11] | H. He, Y. Bai, E. A. Garcia, S. Li, ADASYN: Adaptive synthetic sampling approach for imbalanced learning, in 2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence), (2008), 1322–1328. https://doi.org/10.1109/IJCNN.2008.4633969 |
[12] | H. Han, W. Wang, B. Mao, Borderline-SMOTE: A new over-sampling method in imbalanced data sets learning, in International Conference on Intelligent Computing, 3644 (2005), 878–887. https://doi.org/10.1007/11538059_91 |
[13] |
G. Douzas, F. Bacao, F. Last, Improving imbalanced learning through a heuristic oversampling method based on k-means and SMOTE, Inf. Sci., 465 (2018), 1–20. https://doi.org/10.1016/j.ins.2018.06.056 doi: 10.1016/j.ins.2018.06.056
![]() |
[14] |
Y. Yan, Y. Jiang, Z. Zheng, C. Yu, Y. Zhang, Y. Zhang, LDAS: Local density-based adaptive sampling for imbalanced data classification, Expert Syst. Appl., 191 (2022), 116213. https://doi.org/10.1016/j.eswa.2021.116213 doi: 10.1016/j.eswa.2021.116213
![]() |
[15] |
Y. Xie, M. Qiu, H. Zhang, L. Peng, Z. Chen, Gaussian distribution based oversampling for imbalanced data classification, IEEE Trans. Knowl. Data Eng., 34 (2022), 667–669. https://doi.org/10.1109/TKDE.2020.2985965 doi: 10.1109/TKDE.2020.2985965
![]() |
[16] | H. Bhagwani, S. Agarwal, A. Kodipalli, R. J. Martis, Targeting class imbalance problem using GAN, in 2021 5th International Conference on Electrical, Electronics, Communication, Computer Technologies and Optimization Techniques (ICEECCOT), (2021), 318–322. https://doi.org/10.1109/ICEECCOT52851.2021.9708011 |
[17] |
S. Maldonado, C. Vairetti, A. Fernandez, F. Herrera, FW-SMOTE: A feature-weighted oversampling approach for imbalanced classification, Pattern Recognit., 124 (2022), 108511. https://doi.org/10.1016/j.patcog.2021.108511 doi: 10.1016/j.patcog.2021.108511
![]() |
[18] |
E. Kaya, S. Korkmaz, M. A. Sahman, A. C. Cinar, DEBOHID: A differential evolution based oversampling approach for highly imbalanced datasets, Expert Syst. Appl., 169 (2021), 794–801. https://doi.org/10.1016/j.eswa.2020.114482 doi: 10.1016/j.eswa.2020.114482
![]() |
[19] |
W. Xie, G. Liang, Z. Dong, B. Tan, B. Zhang, An improved oversampling algorithm based on the samples' selection strategy for classifying imbalanced data, Math. Probl. Eng., 2019 (2019), 3526539. https://doi.org/10.1155/2019/3526539 doi: 10.1155/2019/3526539
![]() |
[20] |
L. Peng, H. Zhang, B. Yang, Y. Chen, A new approach for imbalanced data classification based on data gravitation, Inf. Sci., 288 (2014), 347–373. https://doi.org/10.1016/j.ins.2014.04.046 doi: 10.1016/j.ins.2014.04.046
![]() |
[21] |
F. Rahmati, H. Nezamabadi-Pour, B. Nikpour, A gravitational density-based mass sharing method for imbalanced data classification, SN Appl. Sci., 2 (2020). https://doi.org/10.1007/s42452-020-2039-2 doi: 10.1007/s42452-020-2039-2
![]() |
[22] |
M. Koziarski, B. Krawczyk, M. Wozniak, Radial-Based oversampling for noisy imbalanced data classification, Neurocomputing, 343 (2019), 19–33. https://doi.org/10.1016/j.neucom.2018.04.089 doi: 10.1016/j.neucom.2018.04.089
![]() |
[23] | C. Bunkhumpornpat, K. Sinapiromsaran, C. Lursinsap, Safe-Level-SMOTE: Safe-Level-Synthetic Minority Over-Sampling TEchnique for handling the class imbalanced problem, in Pacific-Asia Conference on Knowledge Discovery and Data Mining, 5476 (2009), 475–482. https://doi.org/10.1007/978-3-642-01307-2_43 |
[24] |
Y. Sun, L. Cai, B. Liao, W. Zhu, J. Xu, A robust oversampling approach for class imbalance problem with small disjuncts, IEEE Trans. Knowl. Data Eng., 35 (2023), 5550–5562. https://doi.org/10.1109/TKDE.2022.3161291 doi: 10.1109/TKDE.2022.3161291
![]() |
[25] |
S. Yin, X. Zhu, C. Jing, Fault detection based on a robust one class support vector machine, Neurocomputing, 145 (2014), 263–268. https://doi.org/10.1016/j.neucom.2014.05.035 doi: 10.1016/j.neucom.2014.05.035
![]() |
[26] |
B. Scholkopf, J. C. Platt, J. Shawe-Taylor, A. J. Smola, R. C. Williamson, Estimating the support of a high-dimensional distribution, Neural Comput., 13 (2001), 1443–1471. https://doi.org/10.1162/089976601750264965 doi: 10.1162/089976601750264965
![]() |
[27] |
R. Barandela, R. M. Valdovinos, J. S. Sánchez, New applications of ensembles of classifiers, Pattern Anal. Appl., 6 (2003), 245–256. https://doi.org/10.1007/s10044-003-0192-z doi: 10.1007/s10044-003-0192-z
![]() |
[28] | C. Li, Classifying imbalanced data using a bagging ensemble variation (BEV), in Proceedings of the 45th annual southeast regional conference, (2007), 203–208. https://doi.org/10.1145/1233341.1233378 |
[29] |
S. Hido, H. Kashima, Y. Takahashi, Roughly balanced bagging for imbalanced data, Stat. Anal. Data Min., 2 (2009), 412–426. https://doi.org/10.1002/sam.10061 doi: 10.1002/sam.10061
![]() |
[30] |
B. Chen, S. Xia, Z. Chen, B. Wang, G. Wang, RSMOTE: A self-adaptive robust SMOTE for imbalanced problems with label noise, Inf. Sci., 553 (2021), 397–428. https://doi.org/10.1016/j.ins.2020.10.013 doi: 10.1016/j.ins.2020.10.013
![]() |
[31] |
H. K. Lee, S. B. Kim, An overlap-sensitive margin classifier for imbalanced and overlapping data, Expert Syst. Appl., 98 (2018), 72–83. https://doi.org/10.1016/j.eswa.2018.01.008 doi: 10.1016/j.eswa.2018.01.008
![]() |
[32] |
J. Lu, A. Liu, F. Dong, F. Gu, J. Gama, G. Zhang, Learning under concept drift: A review, IEEE Trans. Knowl. Data Eng., 31 (2019), 2346–2363. https://doi.org/10.1109/TKDE.2018.2876857 doi: 10.1109/TKDE.2018.2876857
![]() |
[33] | M. K. Paul, B. Pal, Gaussian mixture based semi supervised boosting for imbalanced data classification, in 2016 2nd International Conference on Electrical, Computer & Telecommunication Engineering (ICECTE), (2016). |
[34] |
Y. Xie, L. Peng, Z. Chen, B. Yang, H. Zhang, H. Zhang, Generative learning for imbalanced data using the Gaussian mixed model, Appl. Soft Comput., 79 (2019), 439–451. https://doi.org/10.1016/j.asoc.2019.03.056 doi: 10.1016/j.asoc.2019.03.056
![]() |
[35] |
A. Shapiro, Monte carlo sampling methods, Handb. Oper. Res. Manage. Sci., 10 (2003), 353–425. https://doi.org/10.1016/S0927-0507(03)10006-0 doi: 10.1002/wics.1314
![]() |
[36] |
D. P. Kroese, T. Brereton, T. Taimre, Z. I. Botev, Why the Monte Carlo method is so important today, WIREs Comput. Stat., 6 (2014), 386–392. https://doi.org/10.1002/wics.1314 doi: 10.1002/wics.1314
![]() |
[37] |
S. Bej, N. Davtyan, M. Wolfien, M. Nassar, O. Wolkenhauer, LoRAS: An oversampling approach for imbalanced datasets, Mach. Learn., 110 (2021), 279–301. https://doi.org/10.1007/s10994-020-05913-4 doi: 10.1007/s10994-020-05913-4
![]() |
![]() |
![]() |