Green AI for Threat Detection:
A Sustainable Approach to
Securing IoT Networks
Table of contents
Research Suggested
01 Topic
03 Methodology
What is our Research Topic Out Suggested Methodology,
and introduction Model Evaluation, Deployment
Our Model
02 Objective
04 Evaluation
We have discussed aboud our
Our objective and goals
Model Evaluation, Deployment
and Comparative Study
Research
01 Objective
Green AI for Threat Detection:
A Sustainable Approach to
Securing IoT Networks
Green AI for Threat Detection:
A Sustainable Approach to
Securing IoT Networks
Green AI for Threat Detection:
A Sustainable Approach to Securing IoT
Networks
Green AI,is a more environmentally friendly approach to AI that aims
to reduce the carbon footprint of AI systems. It uses energy-efficient
algorithms and models to reduce the amount of energy consumed by
AI systems
Green AI for Threat Detection:
A Sustainable Approach to Securing IoT
Networks
THREATS
Physical Threats
Psychological Threats
Social Threats
Economic Threats
cyber security Threats
Green AI for Threat Detection:
A Sustainable Approach to Securing IoT
Networks
The process of identifying potential security
risks to organizational and personal digital
assets
Introduction
Threats of IoT Network
Distributed Denial-of-Service
Harnessing the power of Green AI, this (DDoS) Attacks
study revolutionizes IoT threat Man-in-the-Middle (MITM)
detection by blending sustainability Attacks
with robust security, ensuring eco-
friendly protection for interconnected Routing Attacks
etc.
networks.
Distributed Denial-of-
Service (DDoS) Attacks
Man-in-the-Middle (MITM)
Attacks
Hacker
User Server
IOT Devices(End devices) Hacker
Routing Attacks
Our
02 Objective
Our Objectives
Develop an energy-efficient AI-based framework
for real-time IoT threat detection without
compromising accuracy.
Implement Green AI principles like model pruning and
quantization to minimize computational costs and
carbon footprint.
Design scalable solutions for seamless deployment in
edge and cloud environments tailored to diverse IoT
ecosystems.
Suggested
03 Methodology
Our Suggested Methodology
Model Design Energy- Evaluation and
Data
Efficiency Deployment
Processing
Optimization
Data Processing
Dataset Preparation: Used an authentic labaled
dataset with class balancing techniques to
comprehensively represent IoT threat scenarios.
Feature Extraction and Normalization: Extract key
features like packet size and protocol type, applying
dimensionality reduction (e.g., PCA) and
normalization to streamline data for efficient
modeling
Dataset Selection and Authenticity
We have used the Aposemat IoT-23
dataset by the Avast AIC laboratory
with the funding of Avast Software.
Dataset link: https://www.stratosphereips.org/datasets-iot23
Dataset Pruning and Quantizing
Data pruning and quantizing are used to enhance energy
efficiency. Pruning removes redundant model weights, reducing
complexity and making the model lightweight, while quantizing
converts high-precision values to lower-precision formats,
significantly reducing memory use and speeding up computations.
These techniques ensure the model operates efficiently on low-
power hardware, aligning with Green AI principles.
Dataset Pruning and Quantizing
Fields for Pruning
id.orig_h, id.resp_h (Origin/Response Host)
local_orig / local_resp (Local Origin/Response)
missed_bytes
tunnel_parents
Fields for Quantizing
Duration
orig_bytes / resp_bytes (Origin/Response Bytes)
orig_pkts / resp_pkts (Origin/Response Packets)
ts (Timestamp)
id.orig_p / id.resp_p (Origin/Response Port)
Dataset Labels
All labels in the dataset
Cleaned
Dataset
Feature Extraction
Feature extraction focuses on key
attributes like packet size,
protocol type, flow duration, and
source/destination ports to
identify patterns of malicious
behavior. By selecting these
relevant features, the process
reduces noise and redundancy,
ensuring efficient and accurate
threat detection.
Model Design
The model design utilizes the
XGBoost Algorithm, integrating key
IoT features like packet size, protocol
type, and flow duration. Preprocessing
includes handling missing data,
encoding categorical features, and
scaling numerical values.
Hyperparameter tuning and stratified
splits ensure optimal performance and
balanced learning, enabling efficient
and accurate threat detection in IoT
networks.
Energy-Efficiency Optimization
This research optimizes energy efficiency by utilizing XGBoost with
selected hyperparameters, including a learning rate of 0.1, a maximum
depth of 6, and 100 estimators, to balance computational cost and
performance. Efficient data preprocessing techniques, such as label
encoding for categorical features, scaling for numerical data, and
handling missing values, streamline input processing. Early stopping
further reduces unnecessary iterations during training. Aligned with
Green AI principles, this approach minimizes resource consumption,
making it suitable for energy-constrained IoT environments.
Model
04 Evaluation
Evaluation and Deployment
Evaluation
Accuracy
Precision
Recall
F1-score
AUC-ROC
Evaluation and Deployment
Evaluation
Accuracy
Precision
Recall
F1-score
AUC-ROC
Evaluation and Deployment
Evaluation
Accuracy
Precision
Recall
F1-score
AUC-ROC
Evaluation and Deployment
Evaluation
Accuracy
Precision
Recall
F1-score
AUC-ROC
Evaluation and Deployment
Evaluation
Accuracy
Precision
Recall
F1-score
AUC-ROC
Evaluation and Deployment
Evaluation
Accuracy
Precision
Recall
F1-score
AUC-ROC
Confusion Matrix
Evaluation and Deployment
Deployment
The IoT threat detection model is containerized with Docker for
portability and deployed on energy-efficient edge devices for real-time
detection with minimal latency. For large-scale applications, we suggest
to use Kubernetes orchestrates cloud deployments to handle high
traffic loads efficiently. Continuous monitoring ensures adaptive
optimization, while secure protocols protect data integrity, enabling a
scalable and robust IoT security framework.
Evaluation and Deployment
Deployment
Comparative Study
Sustainable Green IoT
Security for the Protection:
Subject Internet of
Things Using Sustainability-
Artificial Driven Machine
This Paper Intelligence Intelligence for
ArchitecturesErr
or! Reference Malware
source not Defense [15]
found.
Sustainable Sustainability-driven
Sustainable IoT malware defense for
The focus of the security for IoT
threat detection IoT
Research using LSTM-
using XGBoost
based AI
XGBoost SVM with
(Gradient LSTM (Deep
Model Used Boosting Learning)
Gaussian
Machine) Kernel
Energy-efficient Limited (focus on
Energy explicitly methods like optimized kernel
Efficiency addressed pruning and selection)
quantization
Why XGBOOST
• XGBoost is the best choice for structured tabular data, given its balance of
accuracy, efficiency, and scalability. It performs well on tabular data with
numerical and categorical fields and is easier to optimize for resource-
constrained environments like IoT.
• LSTM would be ideal only if the data has significant temporal dependencies
in the ts field, and computational resources are not a constraint.
• SVM with Gaussian Kernel is less suited for this scenario due to its
scalability challenges and computational inefficiency with large or high-
dimensional datasets.
Summary
Model Performance
Accuracy
99.80%
(%)
Precision
99.58%
(%) 99.80% 99.58% 98.74%
Recall Accuracy Precision
Recall
98.74% (%) (%)
(%) (%)
F1-Score
99.18%
(%)
99.18% 99.99%
F1-Score AUC ROC
(%) Score
THANK YOU
