A Hybrid Semi-Supervised Anomaly Detection Model for High-Dimensional Data

Algorithm 1

The procedure of the proposed hybrid model.

Input: Training set , query sample , desired false alarm rate , nearest neighbor ,

number of subsets , number of elements in one subset , hyperparameter

sets    for DAE.

Output: The label of query sample

(1) Train DAE with the whole training set and predefined parameter set .

(2) Process and with the trained DAE. The dimension-reduced training set is

denoted as .

(3) Generating subsets .

(4) for    do

(5) for    do

(6) for    do

(7) Calculate the th nearest neighborhood distance for in subset .

(8) end for  

(9) Calculate with Eq. (8).

(10) end for

(11) end for

(12) Repeat (6)–(9) to calculate .

(13) for    do

(14)Calculate with Eq. (9).

(15)Calculate with Eq. (10).

(16) end for

(17) Calculate the final decision using Eq. (11).