IoT-Guard: AI-Powered Intrusion Detection for

Securing IoT Networks

K Murugesh
Computer Science and Engineering Madhumithan S Mahesh R
Dayananda Sagar University Computer Science and Engineering Computer Science and Engineering
Ramanagara, India Dayananda Sagar University Dayananda Sagar University
[email protected] Ramanagara, India Ramanagara, India
[email protected] [email protected]

ABSTRACT threats. As a result, IoT networks are increasingly targeted

by attackers exploiting these weaknesses.
The IoT-Guard project aims to develop an AI-powered
intrusion detection system (IDS) to enhance the security of IoT
networks, which are increasingly susceptible to cyber threats
such as Distributed Denial of Service (DDoS) attacks, Denial Current traditional security measures—such as firewalls,
of Service (DoS) attacks, bot attacks, and brute force attempts. antivirus software, and intrusion detection systems—are
With the rapid expansion of IoT devices, traditional security often ineffective in detecting or predicting advanced and
measures often fall short in addressing the unique challenges evolving cyber threats that specifically target IoT
of these networks. IoT-Guard will leverage machine learning
environments. For example, attacks like Distributed
algorithms to analyze IoT network logs and detect anomalies
that could indicate potential intrusions or malicious activities. Denial of Service (DDoS), Denial of Service (DoS), brute
By using historical data, the system will train models to force, and botnet attacks are becoming increasingly
identify patterns and threats associated with various attack prevalent. DDoS and DoS attacks overwhelm the IoT
vectors, such as unusual traffic spikes from DDoS or DoS network by flooding it with traffic, rendering the system
attacks, bot-driven behavior, and unauthorized login attempts unusable. Brute force attacks target the weak
through brute force. The solution will enable the detection of authentication systems of IoT devices, attempting to guess
these attacks early, before they escalate into significant passwords or encryption keys through sheer trial and
security breaches.
error. Botnet attacks involve a network of compromised
IoT devices that are used to carry out large-scale attacks,
often without the knowledge of the device owners.
IoT-Guard’s system will continuously improve through
predictive analytics and real-time analysis, offering proactive
protection to IoT networks against evolving cyber threats. By
monitoring and responding to threats in real time, it can In response to these evolving threats, the IoT-Guard
mitigate risks like botnet-driven DDoS attacks or brute force project aims to develop an AI-powered intrusion detection
intrusions. Ultimately, IoT-Guard seeks to provide a robust, system (IDS) that leverages advanced machine learning
intelligent security solution that can predict and mitigate risks, algorithms to enhance the security of IoT networks.
enhancing the overall safety and reliability of IoT Unlike traditional approaches, IoT-Guard uses AI and
environments
machine learning models to analyze vast amounts of IoT
.Keywords - IoT-Guard, AI-powered intrusion detection network logs, identifying suspicious activity that could
system, IoT networks, machine learning, anomaly detection, indicate an attack. The system is designed to not only
network logs, cybersecurity, predictive analytics, IoT data. detect known threats but also predict potential attacks
before they occur by identifying patterns and anomalies in
network behavior.
I. INTRODUCTION

The rapid growth of the Internet of Things (IoT) has

revolutionized various sectors by connecting millions of To effectively identify and mitigate these threats, the IoT-
devices, from household appliances to industrial Guard system will utilize robust machine learning
equipment. However, this widespread adoption has techniques such as voting classifiers and random forests.
introduced significant security risks, as many IoT devices Voting classifiers, which combine multiple models to
suffer from inadequate built-in protections and insufficient make more accurate predictions, will be used to aggregate
security measures. These devices, often with limited the output of different machine learning models,
computational power, are vulnerable to a variety of cyber improving the overall accuracy of threat detection. This

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE

ensemble method ensures that the system is not overly Jiang Hongyu et al. (2020) focused on the application of
reliant on a single model and can handle the diverse range Gated Recurrent Unit (GRU) Neural Networks to analyze
of IoT devices and potential attacks. On the other hand, and predict emoji usage in video comments, a novel
random forests, a type of decision tree algorithm, will help approach that could be instrumental in understanding user
the system identify complex patterns and relationships in interactions within IoT environments. Their study aimed
the data by creating a collection of decision trees, each to predict the category and position of emojis within video
making a prediction based on a random subset of the data. comments, offering insights into user engagement and
This allows the system to make more accurate predictions, sentiment analysis.
even in the presence of noisy or incomplete data, which is
often the case in IoT networks. Li Y. et al. (2018) demonstrated the effectiveness of deep
learning in detecting anomalies within industrial IoT
systems, achieving high accuracy rates. This research,
published in the IEEE Transactions on Industrial
By training on historical IoT network data and Informatics, is crucial for the development of IDS capable
continuously updating its models, IoT-Guard will not only of adapting to the complex and dynamic nature of IoT
be able to detect anomalies but also predict potential networks.
intrusions in real-time, offering proactive protection to
IoT environments. The goal is to provide a robust, Smith J. et al. (2019) highlighted the importance of real-
intelligent security solution that can prevent attacks like time threat detection and mitigation in IoT security,
DDoS, DoS, brute force, and botnet attacks before they aligning with the goals of the IoT-Guard project. Their
cause significant damage, ultimately enhancing the overall work, presented at the ACM Workshop on Security and
security of IoT networks and ensuring the continued Artificial Intelligence, emphasized the need for proactive
growth and trust in IoT technologies. security measures that can swiftly respond to emerging
threats.

Zhang W. et al. (2020) showed that ensemble learning can

significantly improve the accuracy and reliability of
intrusion detection systems. Published in the IEEE
Internet of Things Journal, their research suggests that
combining multiple learning models can lead to more
robust security solutions, beneficial for the IoT-Guard
project.

Kim S. et al. (2021) explored the use of reinforcement

learning to optimize security policies in IoT networks,
helping to adapt to evolving threats. This study, presented
at the International Conference on Information and
Communications Technology, offers a flexible and
responsive security strategy that could enhance the IoT-
Guard project's ability to counteract new and sophisticated
attacks.
II. LITERATURE REVIEW III. DESIGN
Iman Sharafaldin et al. (2018) proposed a methodical System Architecture Overview:
approach to creating a reliable intrusion detection system
(IDS) dataset, which is crucial for evaluating and 1. Input Layer: Data Collection
enhancing network security measures. By employing
The first layer of the system is responsible for gathering
CICFlowMeter for feature extraction and leveraging a
comprehensive data from various sources within the IoT
suite of machine learning algorithms including
network. This includes collecting network logs that
RandomForestRegressor, KNN, RF, ID3, Adaboost, MLP,
encompass traffic data, device interactions, and system
Naive-Bayes, and QDA, they successfully generated a
events. The logs are raw and unprocessed, capturing the
new IDS dataset named CICIDS2017, comprising 80
essence of the network's activities which may include both
distinct features. This dataset was thoroughly evaluated
benign and malicious behaviors. This data is essential for
using the aforementioned algorithms, demonstrating the
the system as it serves as the foundation for identifying
significance of careful feature selection and algorithm
and understanding the patterns of network behavior. The
choice in achieving effective intrusion detection. This
Input Layer ensures that all relevant data is captured in
research is highly relevant to our project as it underscores
real-time and forwarded to the next layer for further
the importance of a robust dataset and analytical methods
analysis.
in developing an efficient IDS, aligning with our
objectives to improve network security through advanced 2. Preprocessing Layer: Data Preparation
machine learning techniques.
Once the data is collected, it moves to the Preprocessing
Layer where it undergoes a series of cleaning and
structuring processes. The primary goal of this layer is to
refine the raw data by removing noise, such as outliers
and irrelevant information, which could skew the analysis
and lead to inaccurate conclusions. This layer also
involves normalizing the data to ensure consistency and
handling missing values to maintain data integrity. The
Preprocessing Layer is crucial as it transforms the raw
logs into a structured format that is suitable for machine
learning algorithms to process effectively. Platforms, Systems, Services, and Processes:

3. Training Layer: Model Development 1. Data Collection Systems:

The Training Layer is where the core machine learning The foundation of the IDS is built upon robust Data
models are developed. Historical data, which has been Collection Systems that are responsible for gathering logs
preprocessed, is used to train these models. The models and traffic data from IoT network devices. These systems
are designed to learn and recognize patterns that include a diverse array of IoT devices, routers, and other
distinguish between normal network behavior and sensors strategically placed throughout the network to
malicious activities. Various machine learning algorithms capture a wide range of activities. The data collected
are employed, and they are trained to identify anomalies encompasses various types of information, such as device
that may indicate security threats. This layer involves communications, user interactions, system events, and
feature selection, where the most relevant features from network traffic patterns. This data is crucial for the IDS as
the data are chosen to improve the model's predictive it provides the raw material for analysis and threat
accuracy. The models are continuously trained and detection. The Data Collection Systems are designed to
updated to adapt to new types of threats and to refine their operate in real-time, ensuring that all network activities
detection capabilities. are continuously monitored and recorded without any
delays or gaps.
4. Prediction Layer: Threat Assessment
2. Machine Learning Libraries/Frameworks:
The Prediction Layer utilizes the trained models to predict
potential threats in real-time. It uses sophisticated To process and analyze the collected data, the IDS
algorithms such as Voting Classification and Random employs advanced Machine Learning
Forest to analyze incoming network logs and compare Libraries/Frameworks. These frameworks, such as
them against the learned patterns from the Training Layer. TensorFlow and scikit-learn, provide the necessary tools
This layer assesses the threat levels by identifying and algorithms for building and training machine learning
deviations from normal behavior and determining the models. The models used in the IDS include sophisticated
likelihood of malicious intent. The Prediction Layer is algorithms like Random Forest and Voting Classification,
designed to provide a high level of accuracy in threat which are specifically chosen for their effectiveness in
detection, minimizing false positives and ensuring that detecting anomalies and predicting threats. These libraries
potential intrusions are flagged promptly. and frameworks are continuously updated to incorporate
the latest advancements in machine learning, ensuring that
5. Output Layer: Alert Generation the IDS remains at the cutting edge of intrusion detection
technology.
The final layer of the system is the Output Layer, which is
responsible for generating alerts for network 3. Data Preprocessing Pipeline:
administrators. Based on the predictions made by the
models in the Prediction Layer, this layer produces The Data Preprocessing Pipeline is a critical component of
actionable insights that indicate potential intrusions or the IDS that ensures the quality and consistency of the
risks. The alerts are designed to be clear and informative, data used for model analysis. This pipeline involves a
providing administrators with the necessary details to series of steps designed to clean and format the IoT logs,
understand the nature of the threat and take appropriate making them suitable for machine learning algorithms.
action. This layer ensures that the system not only detects The preprocessing includes tasks such as removing noise,
threats but also communicates them effectively to enable handling missing values, normalizing data, and extracting
timely response and mitigation. relevant features. The Data Preprocessing Pipeline is
essential for improving the accuracy of the machine
learning models by ensuring that the data they analyze is
clean, structured, and relevant to the task of intrusion
detection.

4. Prediction and Alert Service:

The Prediction and Alert Service is the final component of 2. Data Collection
the IDS that translates the predictions made by the
machine learning models into actionable insights for users Device Instrumentation: Deploy IoT devices, routers, and
and administrators. This service uses the output from the sensors to collect network logs and traffic data.
models to identify potential threats and risks within the Data Sources Identification: Identify and access various
IoT network. Once a threat is detected, the Prediction and data sources, such as system logs, network traffic, and
Alert Service generates alerts that are sent to users and device interactions.
administrators through a notification system or a user
dashboard. These alerts provide detailed information Data Collection Tools: Utilize tools and scripts to
about the nature of the threat, its potential impact, and automate the collection of data from diverse sources.
recommended actions to mitigate the risk. The Prediction
3. Data Preprocessing
and Alert Service is designed to be user-friendly and
accessible, ensuring that users can quickly understand and Data Cleaning: Remove noise, outliers, and irrelevant
respond to the alerts, thereby enhancing the overall information from the collected data.
security of the IoT network.
Data Normalization: Standardize the data to ensure
consistency across different devices and network
protocols.

Feature Extraction: Identify and extract relevant features

from the data that can be used for intrusion detection.

Data Labeling: Label the preprocessed data as normal or

malicious based on known attack patterns and expert
analysis.

4. Model Selection and Training

Algorithm Selection: Choose appropriate machine

learning algorithms, such as Random Forest, Voting
Classification, and Neural Networks.

Library/Framework Selection: Select machine learning

libraries or frameworks, such as TensorFlow or scikit-
learn, for building and training models.

Model Training: Train the selected models on the

preprocessed and labeled data using historical attack data.

Model Validation: Validate the trained models using a

separate validation dataset to evaluate their performance
and accuracy.

IV. METHODOLOGY

1. Project Planning and Requirements Analysis

Objective Definition: Clearly define the goals of the IDS,

such as detecting various types of cyber threats in IoT
networks.

Requirement Analysis: Identify the functional and non- 5. Model Evaluation and Optimization
functional requirements, including performance metrics,
security standards, and compliance needs. Performance Metrics: Evaluate the models using
performance metrics, such as accuracy, precision, recall,
Scope Determination: Define the scope of the project, and F1-score.
including the types of IoT devices and network protocols
to be supported.
Hyperparameter Tuning: Optimize the models by tuning 10. Project Review and Future Work
hyperparameters to improve their detection capabilities.
Project Review: Conduct a comprehensive review of the
Cross-Validation: Perform cross-validation to ensure the project to assess its success and identify areas for
models generalize well to new, unseen data. improvement.

Ensemble Learning: Combine multiple models using Future Work: Plan for future enhancements, such as
ensemble techniques to improve the overall prediction incorporating new machine learning algorithms,
accuracy. expanding the system's capabilities, and addressing
emerging threats.
6. Prediction and Alert Generation

Real-Time Data Analysis: Analyze real-time network logs

using the trained models to detect potential threats.

Threat Level Assessment: Assess the threat levels based

on the model predictions and assign appropriate risk
scores.

Alert Generation: Generate alerts for administrators when

potential intrusions or risks are detected.

Notification System: Implement a notification system or

user dashboard to communicate alerts to users and
administrators.

7. System Integration and Deployment

System Integration: Integrate the IDS with existing

network infrastructure and security systems.

Deployment Strategy: Develop a deployment plan to roll

out the IDS across the IoT network.

Monitoring and Maintenance: Establish monitoring and

maintenance procedures to ensure the IDS operates
smoothly and remains up-to-date.

8. Testing and Evaluation

Test Environment Setup: Create a controlled test

environment to evaluate the IDS in a realistic setting.

Security Testing: Perform security testing, including V. RESULT

penetration testing and vulnerability assessments, to
validate the IDS's effectiveness.

Performance Evaluation: Assess the IDS's performance in Documentation: Document the system architecture,
terms of detection accuracy, response time, and resource design, and operational procedures.
utilization.
User Training: Provide training to users and
Feedback Loop: Establish a feedback loop to collect user administrators on how to use the IDS and respond to
feedback and system performance data for continuous alerts.
improvement.
Maintenance Guide: Develop a maintenance guide to
9. Documentation and Training assist with system updates and troubleshooting.
Documentation: Document the system architecture,
design, and operational procedures.
VI. CONCLUSION
User Training: Provide training to users and
administrators on how to use the IDS and respond to This study illustrates the important struggles associated
alerts. with early glaucoma detection and the restraints of typical
tests specifically standard automated perimetry and
Maintenance Guide: Develop a maintenance guide to tonometry (modes of ocular diagnostics). Presenting that a
assist with system updates and troubleshooting. large amount of glaucoma cases remain undetected, there
is a need for improved screening methods to adequately
treat glaucoma. The results of this research showed the [6]I. Sharafaldin, A. Habibi Lashkari, and A. A. Ghorbani,
important use of contrast sensitivity tests, which have "IDS 2018 Intrusion CSVs (CSE-CIC-IDS2018)," Kaggle,
been shown to have the highest sensitivity in detecting 2021. [Online]. Available: IDS 2018 Intrusion CSVs.
early glaucomatous damage before vision is significantly
compromised. Throughout the results of the research, it is
suggested that functional/vision tests, specifically contrast A. Some Common Mistakes
sensitivity and motion detection tests, can substantially
improve the likelihood of detecting glaucoma in its early • The word “data” is plural, not singular.
stages. Higher-diagnostic tools utilizing AI-based analysis
and harnessing enhanced imaging tools may help close the • The subscript for the permeability of vacuum 0,
gap in presently accepted assessments. Future studies need and other common scientific constants, is zero with
to pursue validation of these techniques, increase subscript formatting, not a lowercase letter “o”.
accessibility to early screening measures, and measure
• In American English, commas, semicolons,
cost-effectiveness of screening to surpass previously used
periods, question and exclamation marks are located
measures. In addressing early identification of disease, we
within quotation marks only when a complete thought or
are reducing the rates of undetected glaucoma and risk of
name is cited, such as a title or full quotation. When
irreversible blindness will be reduced.
quotation marks are used, instead of a bold or italic
ACKNOWLEDGEMENT typeface, to highlight a word or phrase, punctuation
should appear outside of the quotation marks. A
We would like to acknowledge the assistance and support parenthetical phrase or statement at the end of a sentence
received that helped in the successful completion of this is punctuated outside of the closing parenthesis (like this).
project. First, we take this opportunity to express our (A parenthetical sentence is punctuated within the
gratitude to the Dayananda Sagar University for providing parentheses.)
us with a great opportunity to pursue our Bachelor’s
degree in this institution. We would like to thank Dr. • A graph within a graph is an “inset”, not an
Udaya Kumar Reddy K R, Dean, School of Engineering “insert”. The word alternatively is preferred to the word
& Technology, Dayananda Sagar University for his “alternately” (unless you really mean something that
constant encouragement and expert advice. It is a matter alternates).
of immense pleasure to express our sincere thanks to Dr.
• Do not use the word “essentially” to mean
Girisha G S, Chairman, Department of Computer Science,
“approximately” or “effectively”.
and Engineering, Dayananda Sagar University, for
providing the right academic guidance that made our task • In your paper title, if the words “that uses” can
possible. We would like to thank our guide Mala B A, accurately replace the word “using”, capitalize the “u”; if
Assistant Professor, Department of Computer Science and not, keep using lower-cased.
Engineering, Dayananda Sagar University, for sparing her
valuable time to extend help in every step of our project, • Be aware of the different meanings of the
which paved the way for smooth progress and the fruitful homophones “affect” and “effect”, “complement” and
culmination of the project. We are also grateful to our “compliment”, “discreet” and “discrete”, “principal” and
family and friends who provided us with every “principle”.
requirement throughout the course.
• Do not confuse “imply” and “infer”.
REFERENCES
• The prefix “non” is not a word; it should be
[1]I. Sharafaldin, A. Habibi Lashkari, and A. A. Ghorbani, joined to the word it modifies, usually without a hyphen.
"Toward Generating a New Intrusion Detection Dataset
• There is no period after the “et” in the Latin
and Intrusion Traffic Characterization," in Proc. 4th Int.
abbreviation “et al.”.
Conf. Inf. Syst. Security Priv. (ICISSP), 2018, pp. 108-
116. [Online]. Available: Canadian Institute for • The abbreviation “i.e.” means “that is”, and the
Cybersecurity Datasets. abbreviation “e.g.” means “for example”.
[2]H. Jiang, et al., "Genre-based Emoji Usage Analysis An excellent style manual for science writers is [7].
and Prediction in Video Comments," in Proc. IEEE Int.
Symp. Dependable, Autonomic Secure Comput. (DASC), VI. USING THE TEMPLATE
2020.
After the text edit has been completed, the paper is ready
[3]Y. Li, et al., "Deep Learning for Anomaly Detection in for the template. Duplicate the template file by using the
Industrial IoT Systems," IEEE Trans. Ind. Informatics, Save As command, and use the naming convention
vol. 14, no. 4, pp. 1931-1939, April 2018. prescribed by your conference for the name of your paper.
In this newly created file, highlight all of the contents and
[4]J. Smith, et al., "Machine Learning for IoT Security," in import your prepared text file. You are now ready to style
Proc. ACM Workshop Secur. Artif. Intell., 2019. your paper; use the scroll down window on the left of the
MS Word Formatting toolbar.
[5]W. Zhang, et al., "Ensemble Learning for Intrusion
Detection in IoT Networks," IEEE Internet Things J., vol. A. Authors and Affiliations
7, no. 8, pp. 6666-6677, Aug. 2020.
The template is designed for, but not limited to, six Table Head Table Column Head
authors. A minimum of one author is required for all
conference articles. Author names should be listed starting Table column subhead Subhead Subhead
from left to right and then moving down to the next line. copy More table copya
This is the author sequence that will be used in future
citations and by indexing services. Names should not be a. Sample of a Table footnote. (Table footnote)
listed in columns nor group by affiliation. Please keep
your affiliations as succinct as possible (for example, do Fig. 1. Example of a figure caption. (figure caption)
not differentiate among departments of the same
Figure Labels: Use 8 point Times New Roman for Figure
organization).
labels. Use words rather than symbols or abbreviations
1) For papers with more than six authors: Add when writing Figure axis labels to avoid confusing the
author names horizontally, moving to a third row if reader. As an example, write the quantity
needed for more than 8 authors. “Magnetization”, or “Magnetization, M”, not just “M”. If
including units in the label, present them within
2) For papers with less than six authors: To change parentheses. Do not label axes only with units. In the
the default, adjust the template as follows. example, write “Magnetization (A/m)” or “Magnetization
{A[m(1)]}”, not just “A/m”. Do not label axes with a ratio
a) Selection: Highlight all author and affiliation of quantities and units. For example, write “Temperature
lines. (K)”, not “Temperature/K”.
b) Change number of columns: Select the Columns ACKNOWLEDGMENT (Heading 5)
icon from the MS Word Standard toolbar and then select
the correct number of columns from the selection palette. The preferred spelling of the word “acknowledgment” in
America is without an “e” after the “g”. Avoid the stilted
c) Deletion: Delete the author and affiliation lines expression “one of us (R. B. G.) thanks ...”. Instead, try
for the extra authors. “R. B. G. thanks...”. Put sponsor acknowledgments in the
unnumbered footnote on the first page.

B. Identify the Headings REFERENCES

Headings, or heads, are organizational devices that guide The template will number citations consecutively within
the reader through your paper. There are two types: brackets [1]. The sentence punctuation follows the bracket
component heads and text heads. [2]. Refer simply to the reference number, as in [3]—do
not use “Ref. [3]” or “reference [3]” except at the
Component heads identify the different components of beginning of a sentence: “Reference [3] was the first ...”
your paper and are not topically subordinate to each other.
Examples include Acknowledgments and References and, Number footnotes separately in superscripts. Place the
for these, the correct style to use is “Heading 5”. Use actual footnote at the bottom of the column in which it
“figure caption” for your Figure captions, and “table was cited. Do not put footnotes in the abstract or reference
head” for your table title. Run-in heads, such as list. Use letters for table footnotes.
“Abstract”, will require you to apply a style (in this case, Unless there are six authors or more give all authors’
italic) in addition to the style provided by the drop down names; do not use “et al.”. Papers that have not been
menu to differentiate the head from the text. published, even if they have been submitted for
Text heads organize the topics on a relational, hierarchical publication, should be cited as “unpublished” [4]. Papers
basis. For example, the paper title is the primary text head that have been accepted for publication should be cited as
because all subsequent material relates and elaborates on “in press” [5]. Capitalize only the first word in a paper
this one topic. If there are two or more sub-topics, the next title, except for proper nouns and element symbols.
level head (uppercase Roman numerals) should be used For papers published in translation journals, please give
and, conversely, if there are not at least two sub-topics, the English citation first, followed by the original foreign-
then no subheads should be introduced. Styles named language citation [6].
“Heading 1”, “Heading 2”, “Heading 3”, and “Heading 4”
are prescribed.
C. Figures and Tables [1] G. Eason, B. Noble, and I. N. Sneddon, “On
certain integrals of Lipschitz-Hankel type involving
a) Positioning Figures and Tables: Place figures products of Bessel functions,” Phil. Trans. Roy. Soc.
and tables at the top and bottom of columns. Avoid London, vol. A247, pp. 529–551, April 1955. (references)
placing them in the middle of columns. Large figures and
tables may span across both columns. Figure captions [2] J. Clerk Maxwell, A Treatise on Electricity and
should be below the figures; table heads should appear Magnetism, 3rd ed., vol. 2. Oxford: Clarendon, 1892,
above the tables. Insert figures and tables after they are pp.68–73.
cited in the text. Use the abbreviation “Fig. 1”, even at the
beginning of a sentence. [3] I. S. Jacobs and C. P. Bean, “Fine particles, thin
films and exchange anisotropy,” in Magnetism, vol. III, G.
TABLE III. TABLE TYPE STYLES
T. Rado and H. Suhl, Eds. New York: Academic, 1963, pp.
271–350.
[4] K. Elissa, “Title of paper if known,” unpublished.
[5] R. Nicole, “Title of paper with only first word
capitalized,” J. Name Stand. Abbrev., in press.
[6] Y. Yorozu, M. Hirano, K. Oka, and Y. Tagawa,
“Electron spectroscopy studies on magneto-optical media
and plastic substrate interface,” IEEE Transl. J. Magn.
Japan, vol. 2, pp. 740–741, August 1987 [Digests 9th
Annual Conf. Magnetics Japan, p. 301, 1982].
[7] M. Young, The Technical Writer’s Handbook.
Mill Valley, CA: University Science, 1989.

IEEE conference templates contain guidance text for

composing and formatting conference papers. Please
ensure that all template text is removed from your
conference paper prior to submission to the conference.
Failure to remove template text from your paper may
result in your paper not being published.