Federated Random-Forest Model Generation for Mobile Intrusion Detection

The detection system is segmented into multiple independent sub-models, each responsible for detecting specific types of intrusion behaviors. Each sub-model can be trained and updated independently on local data, then aggregated to form a comprehensive target model. This segmentation enables high detection accuracy across diverse attack types while keeping individual model components relatively simple.

Multiple sub-models from different devices or systems are merged through federated learning to create a unified target model. The merging process combines the strengths of individual sub-models without requiring centralized data collection, achieving superior detection accuracy while maintaining system architecture simplicity through distributed computation.

The training process extracts computational tasks from a centralized location and distributes them to multiple edge devices. Each device performs local training on its own data without exposing raw data or model parameters to a central server. This extraction of training operations to distributed devices eliminates the centralized data collection vulnerability while maintaining training effectiveness through local computation.

An intermediary aggregation mechanism coordinates the federated learning process by collecting model updates from multiple devices and generating the target model without accessing individual device data. This intermediary layer enables secure collaboration between devices, achieving reliable security against label inference attacks while managing training process complexity through automated aggregation operations.