垃圾邮件分类.pptxVIP

垃圾邮件分类器垃圾邮件数据 library(kernlab) data(spam)一共4601封电子邮件，57维预测变量，1维响应变量。makeaddressallnum3douroverremoveinternetordermailreceivewillpeoplereportaddressesfreebusinessemailyoucredityourfontnum000moneyhphplgeorgenum650lablabstelnetnum857datanum415num85technologynum1999partspmdirectcsmeetingoriginalprojectreedutableconference;（）[]!$#capitalAvecaptalLongcapitalTotal贝叶斯公式?定义二值变量表示邮件的类型随机变量表示某个特征，可以是离散型，也可以是连续型。分类规则：以特征“make”为例?是一个连续变量，表示“make”在一封邮件中出现的频率, 取值范围。 head(spam[, ‘make’) [1] 0.00 0.21 0.06 0.00 0.00 0.00 range(spam[, ‘make’]) [1] 0.00 4.54连续变量离散化Usagecut(x, ...) ## Default S3 method: cut(x, breaks, labels = NULL, include.lowest = FALSE, right = TRUE, dig.lab = 3, ordered_result = FALSE, ...)[1] (-0.00454,0.908] (-0.00454,0.908] (-0.00454,0.908] (-0.00454,0.908] (-0.00454,0.908] (-0.00454,0.908]Levels: (-0.00454,0.908] (0.908,1.82] (1.82,2.72] (2.72,3.63] (3.63,4.54]?cut命令将分割为5个区间，则表示‘make’词频的区间。 head(cut(spam[, ‘make’], 5))频率方法?基本思路：分别统计垃圾邮件和非垃圾邮件中‘make’词频的分布列。 table() / sum(table())邮件类型预测真实标签预测标签非垃圾邮件垃圾邮件z=0261189z=164??判别规则第一次小作业使用R实现基于贝叶斯公式的垃圾邮件分类。对57个特征的预测性能进行评价。程序：library(kernlab)library(cowplot)library(ggpubr)old.dir - getwd()setwd(C:/Users/22007/Desktop/spam)# 加载数据data(spam)accuracy_vector-vector(length=57)accuracy_vector[1:57]-0for(iter in 1:57){# 一维特征f - spam[, iter]f - cut(f, 5)# 邮件标签s - spam[, type]# 将数据分为两部分，一部分用于训练模型，一部分用于测试模型train - sample(1:nrow(spam), round(0.9*nrow(spam)))test - setdiff(1:nrow(spam), train)train_f - f[train]train_s - s[train]test_f - f[test]test_s - s[test]# 分布列P(f | s=1)p_f_s1 - table(train_f[train_s==spam]) + 1p_f_s1 - p_f_s1 / sum(p_f_s1))# 分布列P(f | s=0)p_f_s0 - table(train_f[train_s!=spam]) + 1p_f_s0 - p_f_s0 / sum(p_f_s0)# 分布列P(f | s=0)和P(f | s=1)的比较df1 - data.frame(interval = levels(train_f), prob = p_f_s1, type = 1)df0 - data.frame(interval = levels(train_f), prob = p_f_s0, type = 0)df - rbind(df0, df1)df$type - factor(df$type)p1 - ggbarplot(df, x = interval, y = prob.Freq, fill = type, position = position_dodge(0.9), xlab =