Kolmogorov-Smirnov test p-values.

Table summarizing the p-values of the statistical tests of the performances resulting from the testing of each one-site model on each of the other sites in turn. The two samples in each test were the internal performances for that site (from 5-fold cross validation) and the performances of that model on each other site. This was done for each possible combination of the three model types (Logistic Regression, SVM using a linear kernel, and Naive Bayes assuming a Bernoulli distribution) and the two datasets (the whole dataset, and that with the positive cases limited to participants with generalised epilepsy). Thus, there are 6 values for each site. The 2-sample Kolmogorov-Smirnov test has the null hypothesis that the two samples are drawn from the same distribution. Statistical insignificance is taken as insufficient evidence to reject this. There were 7 significant tests for the whole dataset and 4 for the generalised seizure dataset. Agincourt and Kilifi each had 5 significant tests, while the other 3 sites each had at least 5 insignificant tests. (DOCX)