SVM Based DDoS Attack Detection in IoT Using Iot-23 Botnet Dataset

Jul 6, 2026

Upload by Rowan

Smart cities networks face severe threats from distributed denial of service (DDoS) attacks that flood targeted servers, services, or networks with excessive traffic. This study trains and evaluates machine learning models—Support Vector Machine, Decision Tree, and Random Forest—to classify DDoS packets using the IoT-23 botnet dataset. Principal Component Analysis (PCA) is applied for feature selection, and results with PCA are compared against results without PCA using accuracy, precision, F1 score, and recall. PCA reduces execution time with fewer features while maintaining comparable performance, and Decision Tree/Random Forest show stronger classification accuracy. Graphs are plotted with Matplotlib.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages7

Reading Time18 min

Introduction

DDoS attack background and botnet concept

Literature review

Method and experimental setup

PCA-based feature selection

Classifiers and evaluation metrics

Results and comparison

With PCA vs without PCA

Execution time and accuracy analysis

2021 Innovations in Power and Advanced Computing Technologies (i-PACT) ©2021 IEEE DOI: 10.1109/I-PACT52855.2021.9696569| 978-1-6654-2691-6/21/$31.00 | 2021 Innovations in Power and Advanced Computing Technologies (i-PACT) SVM Based DDoS Attack Detection in IoT Using Iot-23 Botnet Dataset D.Nanthiya P.Keerthika S.B.Gopal Department of Computer TechnologyUG Kongu Engineering College Erode, India [nanthiyaratha@gmail.com](nanthiyaratha@gmail.com) S.B.Kayalvizhi Department of Electrical and Electronics Engineering Vivekanandha College of Engineering Namakkal, India [sbkayalvizhi@gmail.com](sbkayalvizhi@gmail.com) Department of Computer Science and Engineering Kongu Engineering College Erode, India [pkeerthika@kongu.ac.in](pkeerthika@kongu.ac.in) T.Raja Department of Electronics and Communication Engineering Vivekanandha College of Engineering Namakkal, India [ecetraja@gmail.com](ecetraja@gmail.com) Department of Electronics and Communication Engineering Kongu Engineering College Erode, India [s.b.gopalece@gmail.com](s.b.gopalece@gmail.com) R.Snega Priya Department of Electronics and Communication Engineering Kongu Engineering College Erode, India [snegapriya2000@gmail.com](snegapriya2000@gmail.com) Abstract—. Network security is one of the most important challenges in Smart cities networks. A distributed denial of service (DDoS) attack is a cyberattack that attempts to distribute the normal traffic of a targeted server, service or network sending flood of Internet traffic. In this paper, Machine learning algorithms such as Support Vector Machine, Decision Tree and Random Forest are trained and tested to classify the DDoS attacked packets. Principal Component Analysis (PCA) is a dimensionality reduction technique that helps to improve the performance of the algorithms. So, in this paper, the efficiency of the PCA is compared with the without PCA results. Initially, the dataset is passed to PCA (Principal Component Analysis) for feature selection and then implemented in different Machine Learning algorithms. The results of all the algorithms are evaluated using the parameters accuracy, precision, F1 score, and Recall. Again, the same dataset is tested and trained in all same ML algorithms using without PCA dataset. Then the results of PCA and without PCA are compared. This experimental analysis shows that by using PCA, the execution time of the algorithm reduces significantly with a smaller number of features and yields same result as that of without PCA. More over the Decision tree and Random Forest algorithms classifies the DDoS packets very accurately compared to SVM. The results are presented as graph using Matplotlib. The dataset taken for our experimental analysis is IoT-23 dataset. Keywords— DDOS, Decision Tree (DT), IOT-23 Dataset, ML, RF, SVM I. INTRODUCTION The aim ofDDoS attacks is to render the website and onlineservice area or network. The goal is to overcome attacks with extra traffic than the server or network. The aim is to provide the online service or network unfeasible. The extra network traffic will accommodate received messages, requests for networkconnections, or forged packets. In some of the cases, battered are endangered by DdoS[9] attack or they are attacked already atan occasional level. The concept based on DDoS attack is very simple and easy, thoughDDoS will direct their level. The fundamental plan in which a DDoS attack might be cyberattack on the server, website,or network flood. If network traffic overcomes target,the server, website or network provided are said to be unworkable. The main method of attack is accomplished which are remotely controlled over the network. This is referred to as “network of bots” or “botnet”. The “network of bots” might direct additional network request than server can maintain or direct maximum collection of data or information which surpass bandwidth abilities of battered victims. Network of bots will be different from lot

FAQs

What is the main goal of the study?

The study aims to detect and classify DDoS attacked packets in IoT networks using machine learning, and to compare results with and without PCA-based feature reduction.

Which machine learning algorithms are evaluated for DDoS packet classification?

Support Vector Machine (SVM), Decision Tree, and Random Forest are trained and tested to classify DDoS packets.

How does PCA affect the experiment results?

PCA is used for feature selection and helps significantly reduce execution time while achieving the same or comparable results compared with using the dataset without PCA.

How are the models evaluated and what tool is used for presenting results?

Evaluation uses accuracy, precision, F1 score, and recall, and the results are presented as graphs using Matplotlib.

Jul 6, 2026

Upload by Rowan