Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Efficient Real Time DDoS Detection Model Using Machine Learning Algorithms

Debashis Kar Suvra

TL;DR

The paper tackles real-time DDoS detection by applying multiple ML classifiers to the CICDDoS2019 dataset, emphasizing a pipeline of data cleaning, StandardScaler normalization, and PCA-based feature selection to balance accuracy with latency. It demonstrates that ensemble methods (Random Forest, AdaBoost, XGBoost) deliver the best accuracy and efficiency, with ROC AUC approaching 0.99+ and fast inference times. A 60/40 train/test split and WEKA-driven PCA reduce feature dimensionality from a large 84+ feature set to a compact representation, enabling effective real-time classification. The authors also implement a real-time client-server setup using Flask and JSON responses, along with visual dashboards, underscoring practical feasibility for live network defense. Overall, the work provides a scalable, high-precision DDoS detection framework with clear pathways for deployment and future enhancement via deep learning and hybrid classifiers.

Abstract

Distributed Denial of Service attacks have become a significant threat to industries and governments leading to substantial financial losses. With the growing reliance on internet services, DDoS attacks can disrupt services by overwhelming servers with false traffic causing downtime and data breaches. Although various detection techniques exist, selecting an effective method remains challenging due to trade-offs between time efficiency and accuracy. This research focuses on developing an efficient real-time DDoS detection system using machine learning algorithms leveraging the UNB CICDDoS2019 dataset including various traffic features. The study aims to classify DDoS and non-DDoS traffic through various ML classifiers including Logistic Regression, K-Nearest Neighbors, Random Forest, Support Vector Machine, Naive Bayes. The dataset is preprocessed through data cleaning, standardization and feature selection techniques using Principal Component Analysis. The research explores the performance of these algorithms in terms of precision, recall and F1-score as well as time complexity to create a reliable system capable of real-time detection and mitigation of DDoS attacks. The findings indicate that RF, AdaBoost and XGBoost outperform other algorithms in accuracy and efficiency, making them ideal candidates for real-time applications.

An Efficient Real Time DDoS Detection Model Using Machine Learning Algorithms

TL;DR

The paper tackles real-time DDoS detection by applying multiple ML classifiers to the CICDDoS2019 dataset, emphasizing a pipeline of data cleaning, StandardScaler normalization, and PCA-based feature selection to balance accuracy with latency. It demonstrates that ensemble methods (Random Forest, AdaBoost, XGBoost) deliver the best accuracy and efficiency, with ROC AUC approaching 0.99+ and fast inference times. A 60/40 train/test split and WEKA-driven PCA reduce feature dimensionality from a large 84+ feature set to a compact representation, enabling effective real-time classification. The authors also implement a real-time client-server setup using Flask and JSON responses, along with visual dashboards, underscoring practical feasibility for live network defense. Overall, the work provides a scalable, high-precision DDoS detection framework with clear pathways for deployment and future enhancement via deep learning and hybrid classifiers.

Abstract

Distributed Denial of Service attacks have become a significant threat to industries and governments leading to substantial financial losses. With the growing reliance on internet services, DDoS attacks can disrupt services by overwhelming servers with false traffic causing downtime and data breaches. Although various detection techniques exist, selecting an effective method remains challenging due to trade-offs between time efficiency and accuracy. This research focuses on developing an efficient real-time DDoS detection system using machine learning algorithms leveraging the UNB CICDDoS2019 dataset including various traffic features. The study aims to classify DDoS and non-DDoS traffic through various ML classifiers including Logistic Regression, K-Nearest Neighbors, Random Forest, Support Vector Machine, Naive Bayes. The dataset is preprocessed through data cleaning, standardization and feature selection techniques using Principal Component Analysis. The research explores the performance of these algorithms in terms of precision, recall and F1-score as well as time complexity to create a reliable system capable of real-time detection and mitigation of DDoS attacks. The findings indicate that RF, AdaBoost and XGBoost outperform other algorithms in accuracy and efficiency, making them ideal candidates for real-time applications.
Paper Structure (22 sections, 14 figures, 3 tables)

This paper contains 22 sections, 14 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Research Objectives
  3. Literature Review
  4. Dataset Analysis
  5. Dataset
  6. Dataset Feature Description
  7. Methodology
  8. Data preprocessing
  9. Feature Selection
  10. Selected Features
  11. Trainning or Testing machine learning classifier
  12. Data visualization
  13. Heat Map
  14. Correlation Matrix
  15. Count Plot
  16. ...and 7 more sections

Figures (14)

  • Figure 1: Workflow Diagram of Prediction model
  • Figure 2: Before data preprocessing with StandardScaler
  • Figure 3: After data preprocessing with StandardScaler
  • Figure 4: PCA for selecting features using WEKA
  • Figure 5: Selected Features Description
  • ...and 9 more figures