處理類別型特徵

R 中的異常偵測入門

Alastair Rushworth

Data Scientist

檢查欄位類別

單一欄位的類別

class(sat$V1)

"numeric"

所有欄位的類別

sapply(X = sat, FUN = class)

      label          V1          V2          V3         V4          V5          V6    high_low
  "numeric"   "numeric"   "numeric"   "numeric"  "numeric"   "numeric"   "numeric" "character"

Isolation forest

將類別特徵編碼為 factor

sat$high_low <- as.factor(sat$high_low)

class(sat$high_low)

"factor"

訓練 isolation forest

sat_for <- iForest(sat[, -1], nt = 100)

含 factor 的 LOF

Gower 距離 可衡量同時含類別與數值特徵的點間距離

library(cluster)
sat_dist <- daisy(sat[, -1], metric = "gower")

將 sat_dist 傳給 lof

sat_lof <- lof(sat_dist, k = 10)

探索 Gower 距離矩陣

物件轉換為矩陣

sat_distmat <- as.matrix(sat_dist)

尋找點間距離的最大與最小值

range(sat_distmat)

0.0000000 0.8680774

一起來練習吧！

R 中的異常偵測入門