Federated learning-based
Intrusion Detection System for
the Internet of Things using
Unsupervised and Supervised
Deep Learning Models
Problem Statement
This paper develops an Federated Learning based Intrusion Detection
System for IoT by combining Unsupervised Autoencoders (AEs) for
Anomaly detection and Supervised Deep Neural Networks (DNNs) for
Attack classification.
Unlike past studies, it integrates both models on a IoT dataset for
improved accuracy and privacy.
Dataset
Name Devices Link
N-BaIoT Home (9) (Doorbell, Webcam, https://archive.ics.uci.edu/dataset/442/detec
Security Camera,Thermostat) tion+of+iot+botnet+attacks+n+baiot
Name Devices Link
Home (105) (Security camera, http://cicresearch.ca/IOTDataset/CIC_IOT_
CIC-IoT2023 Dataset2023/Dataset/
Bulb,Plug, Water Leak Sensor)
Medical (40) (Oxygen Monitor, http://cicresearch.ca/IOTDataset/CICIoMT2
CIC-IoMT2024 024/Dataset/
Heart Rate Monitor)
ACI-IoT Network https://ieee-dataport.org/documents/aci-iot-
Traffic Dataset 23 network-traffic-dataset-2023
http://cicresearch.ca/IOTDataset/CIC%20Io
CIC-IoT DIAD2024 T-IDAD%20Dataset%202024/Dataset/Devic
e%20Identification_Anomaly%20Detection
%20-%20Packet%20Based%20Features/
Basis of comparison
Model Name Accuracy (%) Precision (%) Recall (%) F1-Score (%)
FL-DNN (Proposed) 90.39 99.99 90.10 93.12
RCNN 98.20 92.27 96.64 94.40
CNN-LSTM 90.88 93.04 91.91 88.00
LSTM-RNN 89.47 88.99 82.87 78.19
Random Forest 99.05 - - -
Logistic Regression 82.56 - - -
SVM 82.45 - - -
Naïve Bayes 60.48 - - -
Future Works
Implement other DL, Compare FL
ML algorithms
algorithm/technique
➔ RNN or GNN
➔ FedProx
➔ AdamW optimizer
➔ Scaffold
➔ Ensemble Learning
1. Random forest + ➔ FedDyn
CNN-LSTM ➔ FedOpt
2. DNN+AE+Rando
➔ FedNova
m Forest