Table 3.

Comparison of anomaly detection methods for different cases of data resolutions: high resolution corresponds to an alphabet size of 90, medium resolution corresponds to an alphabet size of 45, and low resolution corresponds to an alphabet size of 10. In this table, $\mu \pm \sigma$ representation is used where $\mu$ is the mean and $\sigma$ is the standard deviation. The proposed PDA method achieves overall best performance (bold entries of table).

	Resolution	PDA	PDD	ZM	CDM	NNS	t-STIDE
ROC AUC	Low	0.948 ±0.011	0.930 ±0.013	0.943 ±0.014	0.787 ±0.017	0.901 ±0.027	0.725 ±0.025
	Medium	0.955 ±0.010	0.943 ±0.011	0.954 ±0.011	0.940 ±0.014	0.918 ±0.022	0.881 ±0.017
	High	0.963 ±0.009	0.959 ±0.009	0.959 ±0.009	0.957 ±0.012	0.920 ±0.021	0.897 ±0.013
PR AUC	Low	0.876 ±0.050	0.871 ±0.052	0.826 ±0.071	0.669 ±0.067	0.719 ±0.098	0.678 ±0.067
	Medium	0.885 ±0.046	0.882 ±0.047	0.881 ±0.053	0.880 ±0.060	0.777 ±0.093	0.828 ±0.050
	High	0.909 ±0.044	0.907 ±0.044	0.895 ± 0.044	0.907 ±0.057	0.777 ±0.091	0.857 ±0.044