Intrusion Classification via Machine Learning Segmentation

The patent segments the intrusion detection task into two distinct stages: first, detecting whether access is unauthorized; second, classifying the specific type of intrusion (first-party, third-party, or synthetic). This segmentation allows each stage to be optimized independently, with the classification stage focusing solely on distinguishing between intrusion types using machine learning models trained on specific parameters like device information, location data, and behavioral patterns.

The patent introduces machine learning models as intermediary components between raw access data and classification decisions. These models process intermediate representations of user behavior, device characteristics, and contextual information to bridge the gap between simple detection and precise classification, enabling accurate intrusion type identification without requiring complex rule-based systems.

The patent implements preliminary action by pre-training machine learning models offline using historical access data and behavioral patterns. During runtime, the pre-trained models quickly classify new access attempts without requiring complex real-time computations. The system also pre-processes and stores relevant parameters (device information, location data, behavioral metrics) to enable rapid inference when classification is needed.

The patent applies partial action by selecting and processing only the most relevant parameters for classification (e.g., device type, location consistency, access patterns) rather than analyzing all possible data points. This selective approach reduces computational overhead while maintaining high classification precision, allowing the system to achieve accurate results without exhaustive processing.

The patent extracts and focuses on specific critical parameters that are most indicative of intrusion types, such as device information (device type, OS version), location data (GPS coordinates, Wi-Fi networks), and behavioral parameters (access frequency, time patterns). By extracting only these essential features rather than processing all available data, the system maintains high detection accuracy while reducing overall data processing complexity.

The patent transforms raw data into meaningful parameters through normalization and feature engineering. For example, location data is converted into distance metrics from known user locations, device information is standardized into comparable formats, and behavioral patterns are aggregated into statistical measures. These parameter transformations make the data more suitable for machine learning analysis while simplifying the processing requirements.