异常值判断处理
1、简单统计量分析
2、3σ准则
data <- rnorm(20)
plot(data,type = "l",lwd=1.5,ylim = c(-4,4),xlim = c(0,23),
xlab=NA,ylab=NA,main = "质量控制图")
lines(rep(mean(data),20),lwd=1.8)
text(21,mean(data),"均值线")
lines(rep(mean(data)-3*sd(data),20),lty=2,col="red",lwd=1.8)
text(21,mean(data)-3*sd(data),labels = "控制下限",col = "red")
lines(rep(mean(data)+3*sd(data),20),lty=2,col="blue",lwd=1.8)
text(21,mean(data)+3*sd(data),labels = "控制上限",col = "blue")
data1 <- read.csv("每日付费及留存数据.csv",header = TRUE)
library(qcc)
attach(data1)
qcc(七日留存率,type="xbar.one",labels=日期,title="新增用户第7日留存率的单值-均值质量监控图",
xlab="data",ylab="第七日留存率")
qcc(data, type, sizes, center, std.dev, limits,
data.name, labels, newdata, newsizes, newdata.name,
newlabels, nsigmas = 3, confidence.level,
rules = shewhart.rules, plot = TRUE, …)
type:
Statistic charted Chart description
“xbar” mean means of a continuous process variable 均值控制图
“R” range ranges of a continuous process variable 均值-极差控制图
“S” standard deviation standard deviations of a continuous variable 均值-标准差控制图
“xbar.one” mean one-at-time data of a continuous process variable 单值-均值控制图
“p” proportion proportion of nonconforming units 可变样本量的转化率
“np” count number of nonconforming units 固定样本量的转化率
“c” count nonconformities per unit 固定样本量的不合格数
“u” count average nonconformities per unit 可变样本量的不合格数
“g” count number of non-events between events 事件间的非事件数
nsigmas = 3 指定用于计算控制限制的σ的数量
3、箱线图分析
boxplot.stats(七日留存率)
idx <- which(七日留存率 %in% boxplot.stats(七日留存率)$out)
boxplot(七日留存率,col="violet")
text(1.1,boxplot.stats(七日留存率)$out,labels=paste(data1[idx,"日期"],
data1[idx,"七日留存率"]),
col="darkgreen")
4、聚类分析
利用多个数值变量
data2<-read.csv("玩家玩牌数据样本.csv")
#归一化处理
data3 <- round(apply(data2[,-1], 2,function(x) (x-min(x))/(max(x)-min(x))),4)
data3 <- data.frame(data3)
row.names(data3) <- data2$用户id
分群
result <- kmeans(data3,3)
centers <- result$centers[result$cluster,]
distances <- sqrt(rowSums((data3-centers)^2))
找出距离最大的
outliers <- order(distances,decreasing = TRUE)[1:5]
print(outliers)
rownames(data3[outliers,])
plot(data3$玩牌局数,data3$正常牌局,pch=result$cluster,axes=FALSE,
xlab="玩牌局数",ylab="正常牌局")
axis(1,labels = F)
axis(2,labels = F)
points(result$centers[,c("玩牌局数","正常牌局")],pch=16,cex=1.5)
points(data3[outliers,c("玩牌局数","正常牌局")],pch="*",col="red",cex=1.5)
text(data3[outliers,c("玩牌局数","正常牌局")],labels=rownames(data3[outliers,]))
