# Kabr Drone Behavior Eval

> Evaluates an automated video classification pipeline for multi-species wildlife behavior monitoring by comparing machine learning predictions against expert manual annotations and traditional ground-based sampling methods. Use when the user wants to benchmark on KABR Drone Behavioral Dataset (custom), or asks about evaluating this task. Reports accuracy.

- Skill: `qhjqhj00/kabr-drone-behavior-eval` (Agent Skill)
- Install (CLI): `npx skillmds add qhjqhj00/kabr-drone-behavior-eval`
- Raw SKILL.md: https://api.skillmd.com/api/skills/qhjqhj00/kabr-drone-behavior-eval/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: DevOps & Infra
- Author: qhjqhj00 (https://skillmd.com/u/qhjqhj00)
- Updated: 2026-09-08
- Page: https://skillmd.com/skills/qhjqhj00/kabr-drone-behavior-eval

---


# kabr-drone-behavior-eval

> kabr-tools: Automated Framework for Multi-Species Behavioral Monitoring — Kline et al. (2025) (arXiv:2510.02030, 2025)

## What this evaluates

Evaluates an automated video classification pipeline for multi-species wildlife behavior monitoring by comparing machine learning predictions against expert manual annotations and traditional ground-based sampling methods.

## Datasets

- **KABR Drone Behavioral Dataset (custom)** — total ?; splits: train (-1), test (-1); repo https://github.com/Imageomics/kabr-tools

## Metrics

- `accuracy` **(primary)** — range: [0, 1]
  - Proportion of correctly classified behavior labels out of total predictions. Computed per class and overall.

## Input / output format

**Input**: Drone video mini-scenes (≥60 seconds) containing multi-species wildlife footage.

**Output**: Discrete behavioral category labels from a predefined set (Graze, Walk, Head Up, Trot, Run, Browse, Auto-Groom, Out of Sight).

## Scoring recipe

```python
def compute_accuracy(predictions, gold):
    correct = sum(1 for p, g in zip(predictions, gold) if p == g)
    return correct / len(gold)

# Confusion matrix computed by counting co-occurrences of (gold, pred) pairs across all test instances.
```

## Common pitfalls

- Visual similarity between behaviors (e.g., Browse vs Graze) causes high misclassification rates.
- Occluded frames are frequently misclassified as Graze due to shared visual features.
- Small sample sizes in case studies (e.g., n=4 for zebra comparison) limit statistical inference.

## Evidence (verbatim from paper)

> The X3D model is evaluated using 25% of the mini-scenes for testing, while the remaining 75% were used for training (Kholiavchenko et al., [2024b]). The confusion matrix (Fig.[6(c)]) compares manual behavior annotations (rows) to predictions generated by the X3D model (columns) across the eight behavioral categories. The X3D model demonstrates high accuracy for core locomotor and foraging behaviors, correctly classifying Graze, Walk, and Head Up with accuracies of 87%, 87%, and 94%, respectively.

## Citation

```bibtex
@misc{kline2025kabrtools,
  title={kabr-tools: Automated Framework for Multi-Species Behavioral Monitoring},
  author={Kline et al. (2025)},
  year={2025},
  note={arXiv:2510.02030}
}
```

- arXiv: 2510.02030