A Deep Random Forest Model on Spark for Network Intrusion Detection

Algorithm 1.

FS-DPRF.

	Input: training dataset D = {(x1, y1), (x2, y2)… (xm, ym)};
	x: potential anomaly data.
	Output: H (x): voting result of sample x;
	CPRF: Deep random forests where {PRFi |i = 1, 2, …, N}.
(1)	CPRF = {∅}
(2)	Initialize hyperparameters: tolerance t and slice window size winSize
(3)	D′ = Feature Grained (D); //D′ is newly generated feature vector.
(4)	do
(5)	i = 1//layer i of cascaded PRF.
(6)	for j = 1, 2, …, T do
(7)	PRFi = {∅}
(8)	D′j ⟵ Bootstrap sampling (D′)
(9)	Treej ⟵ decision tree (D′j)
(10)	PRFi+ = {Treej}
(11)	end for
(12)	if (tolerance ≥ t)
(13)	CPRF+ = {PRFi}
(14)	else
(15)	Break
(16)	i = i + 1
(17)	while (TRUE)
(18)	H (x) = voting method (x)//the last layer votes for classification
(19)	Return CPRF