Neural Network Models of Anti-predator Behavior Thresholds in Birds Exposed to Variable Human Disturbance Regimes

Abstract

The anthropogenic perturbation is gradually defining the avian behavioral scenarios, thus the need to
move past the models of cognitive stasis to those that exemplify cognitive plasticity. This paper will
apply the Artificial Neural Networks (ANNs) to simulate changes in the anti-predator decision-making
threshold of bird populations under varying human disturbance regimes. The correlation between the
frequency of encountering human beings and Flight Initiation Distance (FID) was studied by simulating
1,000 generations of a feed-forward neural architecture. It has been shown via simulations that birds in
High-Frequency or High-Predictability regimes (urban analogues) developed much greater tolerance
thresholds, a reduction in FID on average of 64% (p < 0.001) compared to the situation in low-disturbance
environments. The neural weight analysis shows that the process of habituation is caused by a sensory
filtering layer, which makes the system less responsive to non-lethal stimuli (which occur frequently).
Conversely, in Stochastic/Low-Frequency regimes (rural analogues), the preferred neural configuration
was risk-averse, meaning the Fids of it were 1.8 times higher than the optimal Fids to ensure that a false
negative incurs a high penalty. Statistical information shows that there is a trade-off between the
effectiveness of foraging and survival, with the population in highly disturbed areas showing a 12% gain
of net energy acquisition by lowering the number of unwarranted flights of False Alarm, even though the
potential risk of predation rose by 2.5%. The sensitivity test run proves that the energetic state is the
major modulator of the neural thresholds, and the starvation risk in this case prevails over the learned
habituation in cases where the body fat reserves decline to less than 15 % of the total body mass. Such
results not only offer a mechanistic basis for predicting avian displacement but also imply that
conservation buffer zones need to be scaled based on a certain cognitive adaptation state of local
populations, and not based on predetermined species-wide scales.