Performance counter for biometrics authentication

In the quest for advancing the field of continuous user authentication, we have meticulously crafted two comprehensive datasets: COUNT-OS-I and COUNT-OS-II, each harboring unique characteristics while sharing a common ground in their utility and design principles. These datasets encompass performance counters extracted from the Windows operating system, offering an intricate tapestry of data vital for evaluating and refining authentication models in real-world scenarios.Both datasets have been generated in real-world settings within public organizations in Brazil, ensuring their applicability and relevance to practical scenarios. Volunteers from diverse professional backgrounds participated in the data collection, contributing to the richness and variability of the data. Furthermore, both datasets were collected at a sample rate of every 5 seconds, providing a dense and detailed view of user interactions and system performance. The commitment to preserving user confidentiality is unwavering across both datasets, with pseudonymization applied meticulously to safeguard individual identities while maintaining data integrity and statistical robustness.The COUNT-OS-I dataset was specifically generated in a real-world scenario to evaluate our work on continuous user authentication. This dataset consist of performance counters extracted from the Windows operating system of 26 computers, representing 26 individual users. The data were collected on the computers of the Information Technology Department of a public organization in Brazil.The participants in this study were volunteers, with aged between 20 and 45 years old, consisting of both males and females. The majority of the participants were systems analysts and software developers who performed their routine work activities. There were no specific restrictions imposed on the tasks that the participants were required to perform during the data collection process.The participants used a variety of software applications as part of their regular work activities. This included web browsers such as Firefox, Chrome, and Edge, developer tools like Eclipse and SQL Developer, office programs such as Microsoft Office Word, Excel, and PowerPoint, as well as chat applications like WhatsApp. It's important to note that the list of applications mentioned is not exhaustive, and participants were not limited to using only these applications.For the COUNT-OS-I dataset, the data collected is based on computers with different characteristics and configurations in terms of hardware, operating system versions, and installed software. This diversity ensures a representative sample of real-world scenarios and allows for a comprehensive evaluation of the authentication model.During the data collection process, each sample was recorded at a frequency of every 5 seconds, capturing system data over a period of approximately 26 hours, on average, for each user. This duration provides sufficient data to analyze user behavior and system performance over an extended period. Each sample in the COUNT-OS-I dataset corresponds to a feature vector comprising 159 attributesThe COUNT-OS-II dataset was utilized to evaluate our work in a real-world setting. This dataset comprises performance counters extracted from the Windows operating system installed on 37 computers. These computers possess identical hardware configurations (CPU, memory, network, disk), operating systems, and software installations. The data collection was conducted within various departments of a public organization in Brazil.The participants in this study (37 users) were voluntary administration assistants who performed various administrative tasks as part of their routine work activities. No restrictions were imposed on the specific tasks they were assigned. The participants commonly utilized programs such as the Chrome browser and office applications like Office Word, Excel, and PowerPoint, in addition to the WhatsApp chat application.The data were collected over six days (approximately 48 hours), with sample collected at a 5-second interval. Each sample corresponds to a feature vector composed of 218 attributes. In this dataset, we also apply pseudonymization to hide users' sensitive information.<br>