Brute-Force Attack Detection using Neural Networks

Project Overview

This project implements a neural network-based Intrusion Detection System (IDS) specialized for brute-force attacks. The model is trained on the CIC-IDS 2018 dataset (02-14-2018 subset) and identifies brute-force attack patterns with high accuracy.

The primary goal is to explore deep learning approaches for network security and build a baseline IDS for brute-force detection.

---

Dataset

• Source: CIC-IDS 2018
• Subset Used: February 14, 2018
• Classes:
  • Benign
  • Brute-force attacks (SSH, FTP)
• Number of Features: 64 numeric features after preprocessing
• Preprocessing Steps:
  • Removed NaNs and infinite values
  • Replaced negative values where appropriate (-1 → 0 in window size columns)
  • Removed duplicates selectively
  • Removed zero-variance columns
  • Combined SSH and FTP brute-force attacks into one class
  • Scaled numeric features using StandardScaler
  • Encoded categorical features (Protocol via one-hot encoding, Dst_Port bucketed and one-hot encoded)

---

Model Architecture

• Type: Random Forest classifier
• Input: 18 Features

Baseline Architecture Example:

---

Training & Evaluation

• Train-Test Split: Stratified 80-20
• Metrics Monitored:
  • Precision
  • Recall
  • F1 score
• Performance:
  Near-perfect scores on validation data:
  • F1 score ≈ 0.98
  • Precision ≈ 0.96
  • Recall ≈ 0.98

---

Usage

1. Clone the repository:

git clone <repo-url>
cd <repo-folder>

pip install -r requirments.txt

# preprocessing.ipynb
# model_training.ipynb