Learning for Routing Path Reconstruction and Network Intrusion Detection

Jieling Li

Keywords: Unmanned aerial vehicle, intrusion detection, broad learning, reinforcement learning

Abstract:

A network intrusion detection system is essential
for safeguarding the integrity and availability of sensitive assets.
Network traffic data contains extensive temporal, spatial, and
statistical information. However, existing research has not sufficiently
incorporated spatial-temporal multi-granularity data features
or explored the mutual reinforcement among different types
of data features. To address this, we propose a framework based
on the Broad Learning System (BLS) that can learn and integrate
features across three granular levels. To comprehensively capture
the nuances of traffic data, we construct three granular feature
datasets reflecting time, space, and data content characteristics.
In this work, we employ fundamental broad learning units to
extract abstract features for each granularity, expressing these
features in distinct feature spaces to enhance them individually.
Additionally, we apply He initialization for feature and enhancement
node weights, optimizing the ReLU activation function and
improving detection accuracy over random weight initialization.
Furthermore, under equivalent configuration conditions, the
training and detection times for our Broad Learning System
are comparable to or shorter than those of standard BLS.

Submission Number: 3