Baskethar Eval

---
name: baskethar-eval
description: Evaluates multimodal human activity recognition capabilities in basketball training scenarios by classifying complex dynamic movements from synchronized physiological, inertial, and video sensor data. Use when the user wants to benchmark on BasketHAR, or asks about evaluating this task. Reports F1-score.
metadata:
  skill_kind: dataset_eval
  source_arxiv: 2604.17065
  bibtex_key: gao2026baskethar
  confidence: high
---

# baskethar-eval

> BasketHAR: A Multimodal Dataset for Human Activity Recognition and Sport Analysis in Basketball Training Scenarios — Gao et al. (2026) (arXiv:2604.17065, 2026)

## What this evaluates

Evaluates multimodal human activity recognition capabilities in basketball training scenarios by classifying complex dynamic movements from synchronized physiological, inertial, and video sensor data.

## Datasets

- **BasketHAR** — total ?; splits: train (-1), test (-1)

## Metrics

- `F1-score` **(primary)** — range: [0, 1]
  - Harmonic mean of precision and recall. Calculated per class, then averaged using macro, weighted, and overall accuracy is also reported.
- `Overall Accuracy` — range: [0, 1]
  - Ratio of correctly classified instances to total instances.

## Input / output format

**Input**: Multimodal sensor streams including IMU signals (acceleration, angular velocity, gyroscope), magnetic field, heart rate, skin temperature, and synchronized video frames.

**Output**: Discrete activity class labels (e.g., dribbling, passing, shooting, walking, running).

## Scoring recipe

```python
def compute_metrics(y_true, y_pred):
    precision = precision_score(y_true, y_pred, average=None)
    recall = recall_score(y_true, y_pred, average=None)
    f1 = f1_score(y_true, y_pred, average=None)
    macro_f1 = f1_score(y_true, y_pred, average='macro')
    weighted_f1 = f1_score(y_true, y_pred, average='weighted')
    accuracy = accuracy_score(y_true, y_pred)
    return {'precision': precision, 'recall': recall, 'f1': f1,
            'macro_f1': macro_f1, 'weighted_f1': weighted_f1, 'accuracy': accuracy}
```

## Common pitfalls

- High similarity between running-related activities (walking, running, dribbling while running) often yields <50% accuracy.
- Frequent co-occurrence of actions like dribbling, passing, and bouncing complicates boundary separation for classifiers.

## Evidence (verbatim from paper)

> The quantitative evaluation metrics for the experiments included precision (P), recall (R), and F1-score for activity recognition. For each method, we calculated the precision, recall, and F1-score for each class, as well as the macro-average, weighted average, and overall accuracy. The experiments are performed on our proposed dataset, which is partitioned into training and testing sets with an 8:2 split.

## Citation

```bibtex
@misc{gao2026baskethar,
  title={BasketHAR: A Multimodal Dataset for Human Activity Recognition and Sport Analysis in Basketball Training Scenarios},
  author={Gao et al. (2026)},
  year={2026},
  note={arXiv:2604.17065}
}
```

- arXiv: 2604.17065