direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Inhalt des Dokuments

Bachelor Thesis: Deep Learning for Intrusion Detection

Title:

Deep Learning for Intrusion Detection

Description:

The goal of this thesis is two-fold. First we introduce the concept of an intrusion detection system, which can be described as any algorithm that aims to detect malicious actions either in a host-environment or in a network-environment. We work to replicate results from thepaper "A Deep Learning Approach for Network Intrusion Detection System" by Niyaz, et al. on a dataset, known as NSL-KDD. NSL-KDD is a commonly used benchmark in this field and consists of 41 different features, extracted from data generated by simulated network traffic. Each record has a label associated with it, which states whether that record represents nor-mal or malicious behavior. The main comparison metric is the classification accuracy. Tests against precision and recall are also performed. The main objective is identifying how using a machine-learning model known as a sparse-autoencoder effects the performance on NSL-KDD against softmax-regression. Experiments are also performed with a stacked auto-encoder - an extension of the former, deep neural networks and principal component analysis. We show that while useful in some cases, a deep-neural network might often offer similar performance. The second focus of this thesis is on the actual implementation of those systems. TensorFlow is one of the well-established frameworks for building machine-learning models. PyTorch is a newer alternative. We present two implementations - one for each framework - which in addition to employing for the above-mentioned experiments, we also use as running examples in describing in detail how TensorFlow and PyTorch work. Further evaluation and comparison based on the experience gained with working with them is presented. Finally we provide a
utility that allows further experiments with both implementations to be performed.

Supervisor: Boris Lorbeer, Tanja Deutsch

Type:  Bachelor Thesis

Duration: 4 months

Zusatzinformationen / Extras

Quick Access:

Schnellnavigation zur Seite über Nummerneingabe

Service-centric Networking
Telekom Innovation Laboratories
TEL 19
Ernst-Reuter-Platz 7
10587 Berlin, Germany
Phone: +49 30 8353 58811
Fax: +49 30 8353 58409