Rule induction from examples for expert systems in mouse behavior

Quinlan's ID3 algorithm has been used to induce, from behavioral sequences observed on mice in day and night conditions, rules linking each activity to the three preceding ones in a given sequence. These rules were then input into two expert systems so that they might be validated on other behavioral sequences. It is shown that the succession of various activities exhibited by the mouse is by no means random, but rather displays a Markovian determinism of the first order during the night and of the second order during the day. Among the seven activities singled out in this study, the method applied here has proven very effective in predicting the occurence of rest, locomotion, and grooming in the nest. It is only moderately effective in the cases of feeding and nest-building and fails with respect to drinking and to grooming out of the nest. These results are discussed in the light of current knowledge about the physiological determinants of these various activities. Lastly, the method employed here is compared with other methods having similar uses.