COUNT-OS-I: Windows Performance Counters for Continuous User Authentication

COUNT-OS-I is a dataset of Windows operating system performance counters collected for continuous user authentication research. It contains data from 26 computers representing individual users, sampled every 5 seconds over approximately 26 hours per user on average. The dataset was created by César Andrade from real-world computer usage in a public organization in Brazil.

Use Cases

• Train models for continuous user authentication based on system performance counter patterns.
• Evaluate the robustness of authentication systems against diverse hardware and software configurations mentioned in the description.
• Analyze temporal user behavior patterns based on high-frequency (5-second) sampling of system activity.

Strengths

• Data was collected in a real-world setting from 26 individual users, enhancing ecological validity.
• Samples were recorded at a high frequency of every 5 seconds, providing dense temporal data.
• Data originates from computers with diverse hardware, OS versions, and software, likely increasing model generalizability.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data reflects a specific geographic and organizational context (Brazilian public sector), which may introduce bias.

Provenance

Source

César Andrade

Collection Method

Performance counters were extracted from Windows OS on volunteer users' computers during routine work activities.

Time Range

Data was collected over approximately 26 hours per user on average.

Geography

Data was collected from a public organization in Brazil.