kNN距離スコアの可視化

Rで学ぶ異常検知入門

Alastair Rushworth

Data Scientist

特徴量スケールの標準化

plot(Width ~ Height, data = furniture)

furniture_scaled <- scale(furniture)

plot(Width ~ Height, data = furniture_scaled)

距離行列

furniture_scaled <- scale(furniture)
furniture_knn    <- get.knn(furniture_scaled, 5)

スコアの算出と付与

furniture$score  <- rowMeans(furniture_knn$nn.dist)

head(furniture, 4)

   Height   Width     score
1 58.7179 56.4663 0.4170000
2 54.6154 59.9279 0.3981695
3 58.7179 66.8510 0.2845042
4 63.8462 56.4663 0.4376807

plot(Width ~ Height, cex = sqrt(score), data = furniture, pch = 20)

Rで学ぶ異常検知入門