A Resource-Aware Edge–Cloud Framework for IoT Intrusion Detection under Traffic Shift

Abstract The gateway of an IoT network has to detect attacks with limited computation and limited uplink traffic. We built REC-IDS (Resource-aware Edge–Cloud IDS), in which a decision tree on the gateway classifies every flow and a deferral gate sends a declared share of the flows to a classifier in the cloud. First, we trained the tree and the cloud model on the training data. Second, we took a small labelled window from the beginning of the test period: its first half adapted the models in the adapted setting, and its second half trained the gate and set its threshold. Third, we tested the system on the remaining traffic. We used CIC-DDoS2019, with training and test data from different days, and CICIoT2023, with the first 70% of each class in time for training and the last 30% for testing. On CIC-DDoS2019 with a frozen gateway, the accuracy of the tree alone was 0.880. The learned gate raised it to 0.937, which is at the level of the cloud model alone (0.934), while 11% of the flows were sent and the uplink traffic fell by 89%. On CICIoT2023 the gain was smaller. At a declared budget of 10%, the accuracy went from 0.797 to 0.810 (cloud model 0.810), and after adapting the models to the labelled window it went from 0.817 to 0.835 (cloud model 0.837). Under a random split, the tree alone reached 0.9988 on CIC-DDoS2019 after the removal of conflicting labels and the cloud added nothing. On CICIoT2023 under a random split, the cloud model added one point of accuracy and the gate recovered it with 10% of the flows sent. The benefit of REC-IDS appeared when the test traffic differed from the training traffic: about 90% less uplink traffic at an accuracy close to that of the cloud model. On CICIoT2023 the learned gate was slightly better than a confidence gate.

Authors

Publication Details

Journal
Science Forum (Journal of Pure and Applied Sciences)
Published
2026-11-13
DOI
https://doi.org/10.70882/vrj2bj45
Primary Topic
Network Security and Intrusion Detection
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Resource-Aware Edge–Cloud Framework for IoT Intrusion Detection under Traffic Shift

Ahmed Adnan Hadi
Science Forum (Journal of Pure and Applied Sciences)
Network Security and Intrusion Detection
article

A Resource-Aware Edge–Cloud Framework for IoT Intrusion Detection under Traffic Shift

Ahmed Adnan Hadi
article en

Abstract

Abstract The gateway of an IoT network has to detect attacks with limited computation and limited uplink traffic. We built REC-IDS (Resource-aware Edge–Cloud IDS), in which a decision tree on the gateway classifies every flow and a deferral gate sends a declared share of the flows to a classifier in the cloud. First, we trained the tree and the cloud model on the training data. Second, we took a small labelled window from the beginning of the test period: its first half adapted the models in the adapted setting, and its second half trained the gate and set its threshold. Third, we tested the system on the remaining traffic. We used CIC-DDoS2019, with training and test data from different days, and CICIoT2023, with the first 70% of each class in time for training and the last 30% for testing. On CIC-DDoS2019 with a frozen gateway, the accuracy of the tree alone was 0.880. The learned gate raised it to 0.937, which is at the level of the cloud model alone (0.934), while 11% of the flows were sent and the uplink traffic fell by 89%. On CICIoT2023 the gain was smaller. At a declared budget of 10%, the accuracy went from 0.797 to 0.810 (cloud model 0.810), and after adapting the models to the labelled window it went from 0.817 to 0.835 (cloud model 0.837). Under a random split, the tree alone reached 0.9988 on CIC-DDoS2019 after the removal of conflicting labels and the cloud added nothing. On CICIoT2023 under a random split, the cloud model added one point of accuracy and the gate recovered it with 10% of the flows sent. The benefit of REC-IDS appeared when the test traffic differed from the training traffic: about 90% less uplink traffic at an accuracy close to that of the cloud model. On CICIoT2023 the learned gate was slightly better than a confidence gate.

Science Forum (Journal of Pure and Applied Sciences)
Network Security and Intrusion Detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon