Table 1.

Window-based change detection algorithms.

Approach	Window size	Window preprocessing	Change score	Change significance test
RuLSIF [8]	Any	Hankel matrix	Probability density ratio estimation with Pearson divergence	Threshold learning in supervised applications. N/A for unsupervised applications
Texture-based [5,6]	Any	Grey-level co-occurrence matrix, texture features	Weighted normalized Euclidean distance	N/A
PCAR [7]	Large	$m\times N$ KL distance permutation matrix	Count of time intervals with significant changes (proportion of permuted KL distances greater than observed window)	N/A
sw-PCAR	Any	$N$ KL distance permutation vector	KL distance	Non-parametric outlier detection based on Boxplot analysis
Virtual classifier [9]	Large	Physical activity features (intra-day and inter-day if window size > 1)	Cross validation prediction accuracy of binary classifier	Hypothesis testing based on prediction accuracy exceeding a threshold

KL = Kullback-Leibler, $m$ = number of time intervals, $N$ = number of permutations.