Figure 5

Conclusive results from the classification of each segmentation (see Table 1). Each plot shows the 95% binomial confidence intervals for the classifiers of each segmentation regarding their agreement on the significant difference between the two animal groups for each strategy and the strategy transitions. Squares indicate the mean of the classifiers; errorbars represent the 95% confidence intervals; the dashed line indicates the threshold of interest (0.5 or 50%). Confidence intervals clearly above 0.5 (or 50%) confirm that there is indeed a significant difference between the the two animal groups on the strategies and the strategy transitions. The table below the plots shows the Friedman test p-values (upper table) and the equivalent Friedman’s chi-square statistic (lower table) for the classification result of the ensemble; in all cases k = 2, control and stress columns. Segmentation configurations are arranged in columns and strategies in rows; each element has the relevant p-value and chi-square statistic and bold cells indicate significant difference, i.e. p-value < 0.05. Abbreviations: Thigmotaxis (TT), Incursion (IC), Scanning (SC), Focused Search (FS), Chaining Response (CR), Self Orienting (SO), Scanning Surroundings (SS), Target Scanning (ST), Strategy Transitions (tr) (refer to Methods for more information on each behavioural strategy). We see that in three cases (Segmentations II, III, IV) the two animal groups show significant differences in the strategies of Thigmotaxis (TT), Incursion (IC) and Chaining Response (CR) and transition between strategies (tr). We can see that while Segmentations II, III, IV agree that there is significant differences on the Thigmotaxis, Incursion, Chaining Response and transitions, Segmentation I fails to capture the significant difference on the Chaining Response because of the lengthy segments which caused this strategy to be overshadowed by other strategies and disappear (refer also to Table 3).