TABLE 2.

Results of significance testing after determining the state sequence with Viterbi analysis

Output	Total Seizures, (N)	x*	y**	Predicted Seizures	S1/S2 Ratio	P-Value
HMM-Viterbi	29	8	0.3329	17	6.1344	2.9525 × 10-8
Algorithm implementation	29	7	0.2963	5	8.4019	0.1898

Values represent the seizure prediction performance of the raw SVM-based seizure detector outputs 1) filtered by HMM-Viterbi decoding vs. 2) the particular implementation of the detector in Gardner et al. (2006). The expected S1/S2 ratio of baseline to detected state for HMM-Viterbi output was calculated by the trained HMM’s transition matrix. The same ratio in the author-implemented case was determined empirically from the data by summation of baseline vs. detected outputs.

^*Minimal number of detected seizures for a type I error of 0.05

^**minimum proportion of detected seizures for a type II error of 0.2.