Network Intrusion Detection System (NIDS) using LSTM

This project implements a Deep Learning model to detect network intrusions using the CIC-IDS-2018 dataset. The model is built with PyTorch and leverages an LSTM architecture to classify network traffic

Key Features

• Large-scale Data Handling: Implemented efficient data loading and memory optimization (downcasting, sampling, and garbage collection) to process massive CSV files within Kaggle's RAM limits.
• Multiclass Classification: Detects multiple attack types, including DDoS (LOIC-HTTP), Brute Force (FTP/SSH), and Benign traffic.
• Deep Learning Pipeline: End-to-end PyTorch implementation including custom Dataset, DataLoader, and LSTM architecture.

Performance

The model achieves high reliability across multiple metrics:

• ROC AUC Score: ~0.9945
• DDoS Recall: 0.95 (High detection rate for active threats)
• Overall Accuracy: 98%

Tech Stack

• Framework: PyTorch
• Data Analysis: Pandas, NumPy, Scikit-learn
• Visualization: Matplotlib, Seaborn

Challenges Addressed

• Memory Management: Handled 30GB+ raw data by optimizing data types and using sampling techniques.
• Class Imbalance: Evaluated performance using Confusion Matrix and ROC AUC instead of just Accuracy to ensure minority attack classes are detected.