An Efficient and Fast Model Reduced Kernel KNN for Human Activity Recognition

Algorithm 2

The working process of K-KNN.

	Require: input data matrix, ; the target value, ; the parameter of K-KNN, K; the number of training data, L; the kernel parameter, .
	Ensure: the forecasting output ().
	Data Separation:
(1)	Data matrix (X) is separated as and by equation (3) and (4), separately;
	Kernel Computation Part:
(2)	Calculate the kernel matrix of training features () by equation (6);
(3)	Calculate the kernel matrix of testing features () by equation (7);
Loop:
(4)	fordo
(5)	fordo
(6)	Calculate the distance between and by equation (1);
(7)	end for
(8)	Sort the distance from smallest to largest value (in ascending order);
(9)	Pick the top K vectors from the sorted collection as an index;
(10)	Set the forecasting the class label based on the most frequent class of processed index.
(11)	end for
	return