Algorithm Benchmark Suite v2.0: Synthetic Dataset for Circular Bias Detection

Algorithm Benchmark Suite v2.0

A comprehensive benchmark dataset for evaluating circular bias detection algorithms in AI evaluation contexts.

Dataset Contents

This archive contains 10 CSV files with synthetic evaluation data across diverse scenarios:

algorithm_benchmark_suite.csv - Main benchmark (20 records, 7 fields)

scenario_*.csv (9 files) - Controlled experiments with varying:

Noise levels (0.05-0.2)

Bias intensities (0.0-0.9)

Temporal configurations (8-20 periods)

Algorithm counts (3-6 algorithms)

Data Schema

Each record includes standardized fields:

time_period: Evaluation period ID (integer)

algorithm: Algorithm identifier (string)

performance: Performance metric [0-1] (float)

constraint_compute: Computational constraint (float)

constraint_memory: Memory constraint in GB (float)

constraint_dataset_size: Training dataset size (integer)

evaluation_protocol: Protocol version (string)

Use Cases

Benchmarking bias detection algorithms

Testing evaluation methodology robustness

Research on algorithmic fairness in temporal domains

Validation of circular reasoning detection frameworks

Reproducibility

All data is fully reproducible using the included generation script:
python data/generate_benchmark_data.py --validate

Associated Software

Framework implementation: https://github.com/hongping-zh/circular-bias-detection

Documentation

Data Dictionary: Comprehensive field specifications

JSON Schema: Formal validation rules

README: Usage examples and guidelines

CHANGELOG: Version history and updates

Citation

Zhang, H. (2024). Algorithm Benchmark Suite v2.0 [Data set]. Zenodo. https://doi.org/10.5281/zenodo.17201032