Hidden Markov models: the best models for forager movements?

Area of Science:

• Movement Ecology
• Behavioral Ecology
• Data Science

Background:

• Inferring behavioral modes from movement patterns is crucial in movement ecology.
• Hidden Markov Models (HMMs) are commonly used but may not fully capture behavioral dynamics.

Purpose of the Study:

• To evaluate Hidden Semi-Markov Models (HSMMs) and discriminative models as alternatives to HMMs for inferring behavioral modes.
• To compare the accuracy of HSMMs, HMMs, and discriminative models using human foraging data.

Main Methods:

• Utilized a dataset of over 200 Peruvian fisherman trips with movement data and observed behavioral modes (fishing, searching, cruising).
• Compared Hidden Markov Models (HMMs), Hidden Semi-Markov Models (HSMMs), Random Forests, Artificial Neural Networks, and Support Vector Machines.
• Employed a cross-validation procedure to assess model performance.

Main Results:

• HSMMs achieved the highest accuracy (80%), significantly outperforming HMMs and discriminative models.
• Simulations indicated that HSMMs approach 100% accuracy with higher temporal resolution data.
• The sequential nature of movement data is critical for accurate behavioral mode inference.

Conclusions:

• HSMMs are highly effective for inferring behavioral modes from movement trajectories.
• The study recommends HSMMs for movement ecology research.
• Hybrid HSMM-discriminative models offer potential for further improvements in inference performance.