Statistics & Data Science MS Candidate
When
1:30 – 2:30 p.m., Today
Where
Title: A Comparison of Statistical Learning Models for Classification Applied to Primate Behavior.
Abstract:
Understanding how primates adjust their behavior in human-modified landscapes is central to their conservation. This thesis compares eight statistical learning models for classifying the behavior of moor macaques in South Sulawesi, Indonesia, using 1,993 scan-sampling observations. Five behavioral categories- Rest, Move, Feed/Forage, Social and Sexual, and Human Directed were predicted using location relative to roads, age, sex, and time of day. Model performance was evaluated using 0–1 loss and three proper scoring rules. LASSO achieved the best overall performance, combining well-calibrated probabilities with a sparse and directly interpretable set of coefficients, with 56% of the coefficients shrunk to zero, and was selected as the final model. Proximity to roads was associated with the largest differences in predicted behavior. Human Directed behavior was negligible in forest interiors but increased to as much as 65% at roadside locations, alongside reductions in foraging, resting, and movement. Time of day showed a smaller, location-dependent effect. These findings indicate that road proximity is associated with substantial shifts in macaque activity budgets, highlighting the ecological significance of primate-human interactions in increasingly anthropogenic landscapes.