1.完整项目描述和程序获取
>面包多安全交易平台:https://mbd.pub/o/bread/Y52ZlZds
>如果链接失效，可以直接打开本站店铺搜索相关店铺：
>如果链接失效，程序调试报错或者项目合作也可以加微信或者QQ联系。
2.部分仿真图预览
3.算法概述
从这三个初始聚类中心开始聚类划分。对于一个待分类的对象，计算它到现有聚类中心的距离，若（这个距离）<（现有各个聚类中心距离的最小值），则将这个待分类对象分到与它相距最近的那一类；如果（这个距离）>（现有各个聚类中心距离的最小值），则这个待分类对象就自成一类，成为一个新的聚类中心，然后对所有对象重新归类。SVM训练样本筛选：假定有正常和攻击两类样本，将两类样本分别用K-means聚类，将这些聚类中心点作为新的样本点。然后在正常类中，计算每个新样本点在异常类中距离自己最近的M（假定M=3）个样本点，从而在异常类中假设找到P个样本点。异常类中的这P个样本点找到在正常类中距离自己最近的M个样本点，假定找到了Q个样本点。此时，这P+Q个样本点就是新选出的SVM训练样本。
4.部分源码
is_continue = 1;
C = [C1;C2;C3];
R = [R1;R2;R3];
Leg = ones(length(Attack_Dat),1);
K = 3;
CNT = 0;
while(is_continue == 1)
CNT = 0;
for i = 1:length(Attack_Dat)
for j = 1:size(C,1)
d(i,j) = func_dis(Attack_Dat(i,:),C(j,:));
end
[VV,II] = min(d(i,:));
if VV <= min(R)
Leg(i) = II;
CNT = CNT + 1;
else
K = K + 1;
Leg(i) = K;
C = [C;Attack_Dat(i,:)];
R = [R;alpha*mean(func_dis(Attack_Dat,Attack_Dat(i,:)))];
end
end
if CNT == length(Attack_Dat)
is_continue = 0;
else
is_continue = 1;
end
end
Tmps = unique(Leg);
if min(Tmps) > 1
Leg = Leg - (min(Tmps)-1);
end
Leg = sort(Leg);
Leg = Leg-1;
%检测率
Right1 = 100*(length(find(Leg == T))/length(Attack_Dat))
%误判率
Error1 = 100*(1-length(find(Leg == T))/length(Attack_Dat))
%对比算法K-means
%对比算法K-means
KK = length(unique(Leg));
opts = statset('Display','final');
X = Attack_Dat;
[idx,ctrs] = kmeans(X,KK,'Distance','city','Replicates',10,'Options',opts);
idx = sort(idx);
%检测率
Right2 = 100*(length(find(idx == T))/length(X))
%误判率
Error2 = 100*(1-length(find(idx == T))/length(X))
attack1 = [Attack_Dat(:,:),Leg];
%步骤3：SVM
%步骤3：SVM
%特征选取时，选取的特征数量所对应的的检测率的仿真图
POS = attack1(1:length(Attack_Dat)/2,:);
NEG = attack1(length(Attack_Dat)/2+1:end,:);
%将数据随机产生训练集合和测试集合
INDD= (1:length(POS));
Attack_Dat_train = [POS(INDD(1:round(2/3*length(POS))),1:end-1);NEG(INDD(1:round(2/3*length(POS))),1:end-1)];
T_train = [POS(INDD(1:round(2/3*length(POS))),end) ;NEG(INDD(1:round(2/3*length(POS))),end)];
Attack_Dat_test = [POS(INDD(round(2/3*length(POS)))+1:end,1:end-1);NEG(INDD(round(2/3*length(POS)))+1:end,1:end-1)];
T_test = [POS(INDD(round(2/3*length(POS)))+1:end,end) ;NEG(INDD(round(2/3*length(POS)))+1:end,end)];
%训练集合
POS_train = Attack_Dat_train(find(T_train==1),:);
NEG_train = Attack_Dat_train(find(T_train==2),:);
%测试集合
POS_test = Attack_Dat_test(find(T_test==1),:);
NEG_test = Attack_Dat_test(find(T_test==2),:);
%随机选择K个样本进行分类，并计算对应的检测率
K = 60;
KK = 2;
Alpha = 0.7;
Indes = 0;
Index = 0;
Indes = Indes + 1;
tic
Index = Index + 1;
IND = randperm(length(POS_train));
POS_train2 = POS_train(IND(1:K),:);
NEG_train2 = NEG_train(IND(1:K),:);
%Kmeans
opts = statset('Display','final');
X = [POS_train2;NEG_train2];
[idx,ctrs] = kmeans(X,KK,'Distance','city','Replicates',10,'Options',opts);
%准备SVM训练分类——SVM训练样本的选择
Q = round(Alpha*size(POS_train2,1));
if Q == 0
Q = 1;
end
P = round(Alpha*size(NEG_train2,1));
if P == 0
P = 1;
end
%正样本
d_pos = func_dis2(POS_train2,ctrs(1,:));
%选择距离最近的M个
[Vpos,Ipos]= sort(d_pos);
POS_train_final = POS_train2(Ipos(1:Q),:);
%负样本
d_neg = func_dis2(NEG_train2,ctrs(2,:));
%选择距离最近的N个
[Vneg,Ineg]= sort(d_neg);
NEG_train_final = NEG_train2(Ipos(1:P),:);
%SVM训练
options = optimset('maxiter',10000);
Ps = [POS_train_final;NEG_train_final];
Ts = [ones(length(POS_train_final),1);2*ones(length(NEG_train_final),1)];
SvmNet = svmtrain(Ps,(Ts),'Kernel_Function','linear','Polyorder',2,'quadprog_opts',options);
%SVM测试
Test_Dat = [POS_test;NEG_test];
Classify = svmclassify(SvmNet,Test_Dat);
%计算检测率
times(Index) = toc;
Right(Index) = length(find((Classify)==T_test))/length(T_test);
clc;
05_016_m
