A Realistic, High-Skew Financial Transaction Query Log for Concurrency and Caching Research (v1.0)

This dataset provides a realistic, high-skew, and noisy query log simulating a financial Online Transaction Processing (OLTP) system. It was created for the ICCPEAS 2025 paper: "Improving Transactional Concurrency in Financial Systems: A Proactive Caching Strategy using LSTM". Standard synthetic datasets for database research are often too simplistic, failing to capture the chaotic nature of real-world workloads. This dataset is designed to be an advanced "testbed" for research in database concurrency, lock contention, and proactive caching strategies. The generator script (generate_dataset.py) models four key real-world phenomena: Workload Skew (Zipf's Law): Uses a Zipfian distribution (a=1.2) to select "hot" users and merchants, ensuring that a small number of entities are responsible for a large percentage of transactions. Temporal Rhythm (Time of Day): Simulates a diurnal (day/night) pattern, altering transaction probabilities to reflect high-volume business hours and low-volume overnight activity. Transactional Integrity (The "Signal"): The log is built from three core transaction patterns: 'Purchase', 'Transfer', and 'Balance Check'. Stochastic Chaos (The "Noise"): To test model robustness, the dataset injects two forms of chaos: Transaction Failures (5% Rate): Read/Write transactions randomly abort mid-way to mimic "insufficient funds" or "account frozen" errors. Stochastic Interruptions (2% Rate): A random read_system_audit_log query is injected into the middle of active transactions, breaking the clean patterns. The dataset is provided as a .csv file with 9 columns (see the related GitHub repository for the full schema). The code that generated this data is also available at the linked repository.