Basic Informations

C.V

Name: Rehab Hosny Mohamed

Phone number: 01157624022

Birth of date: 23 / 10 / 1994

Email: rehab.hosne@fcis.bsu.edu.eg

Address: East of the Nile - Beni Suef - Egypt

Master Title

Building an ensemble Multis???? ?????? ???? ??? ?????? ????? ??????? ???????? ?????? ?????? ????????? ?????? ???? ??? ?????? ????? ??????? ???????? ?????? ?????? ????????? ?????? ???? ??? ?????? ????? ??????? ???????? ?????? ?????? ????????? ?????? ???? ??? ?????? ????? ??????? ???????? ?????? ?????? ?????tage Intrusion Detection System using Machine Learning Techniques

Master Abstract

Cyber-attacks have increased these days and become more dangerous in the network, as there are newest attacks such as Ransomware, Denial of Service (DoS), Distributed Denial of Services (DDoS), Man in the middle attack (MITM), Password attack, Cross site scribting attack (XSS) and other types of attacks that can harm the network services and network devices. In these days a lot of devices can connect to each other through IoT networks, so we need an efficient IDS to detect attacks. There are several intrusion detection systems (IDS) that are available for analyzing and predicting network anomalies and threats, but some of them can’t identify the newest attacks and their accuracy remains an issue. In this thesis, we build an IDS to detect the attacks in IoT networks. We use the TON-IoT Windows 10 dataset to develop the system. The ReliefF feature selection algorithm is used to select the most important attributes from the dataset. We do a comparative study to compare our work with previous that select the features from TON-IoT Windows 10 dataset using the correlation function. The machine learning and deep learning techniques are applied to the selected features by ReliefF algorithm and correlation function. The experiment results show that the Medium Neural Network model, Weighted KNN model, and Fine Gaussian SVM model have achieved best results that have an accuracy of 98.39 %, 98.22 %, and 97.97 %, respectively. The LSTM model has achieved low accuracy 70 %, so we use the pre-trained models (ResNet50 and EfficientNet models) to enhance the accuracy of LSTM model, and also tune the Light Gradient Boosting Machine (lightgbm), Random Forest Classifier (rf), and Decision Tree Classifier by changing in the values of hyper-parameters of each models. The ResNet50 and EfficientNet models have achieved high accuracy than LSTM models. And also the Random forest and Light Gradient Boosting Machine have achieved best results.

PHD Title

PHD Abstract