Fig 4. Computers can recognize additional feeding states.

Hidden Markov Models (HMMs) of insect feeding states. (A) Bayesian information criterion (BIC) for HMMs of different numbers of feeding states. BIC conservatively penalizes the likelihood function with increasing numbers of feeding states. Minimum BIC scores indicate a more appropriate number of feeding states; the decreasing BIC scores suggest that the model can resolve more feeding states than the six currently recognized. (B) Three and half hour sample of electrical penetration graph recordings from Asian citrus psyllid on Carrizo citrange citrus. (C) Human-annotated insect feeding states from visual inspection of (B) on a second by second basis. (D) Feeding states recovered from an eight state Hidden Markov Model. The model resolves phloem feeding states E1 and E2 in accordance with human annotation and recognizes more feeding states within the human annotated C feeding state (dashed box in (C) and (D).