Research & Development

Publications:

DDoS Attack Detection in IoT Networks Using Deep Learning Models Combined with Random Forest as Feature Selector

Abstract

Due to the major advancements and finesse earned in technology, the internet and communication fields have seen a ground-breaking breakthrough by incorporating themselves with “things” which as a result have created connected systems and have given multiple utilities to a single device. But due to the digitization of technologies and always online and connected features, they are becoming more prone to cyberattacks, mostly due to the widespread Distributed Denial of Services (DDoS) attacks. Since DDoS attacks create multiple agents to attack a victim network, the always-connected Internet of Things (IoT) devices that communicate over various communication protocols gives the perfect platform for such attacks and these sort of the attacks are on the rise day by day. So, the advent of an efficient and highly precise DDoS attack detection model needs to be addressed. Being motivated by the increasing amount of DDoS attacks, the vulnerability of IoT devices to such attacks, and to tackle this situation, two deep learning-based models have been proposed. The models are based on the combination of Random Forest as a feature selector and 1D Convolutional Neural Network and Multilayer Perceptron methods for DDoS attack detection. The main objective behind the proposed models is to detect DDoS attacks accurately and as early as possible. The models have been evaluated using the CICIDS2017 dataset which resulted in high accuracy of 99.63% in the case of RF-1DCNN and 99.58% in the RF-MLP model.

Conference:

2nd International Conference On Advances In Cyber Security, Universiti Sains Malaysia

Authors: Minhaz Bin Farukee, M. S. Zaman Shabit, Md. Rakibul Haque & A. H. M. Sarowar Sattar

Publisher: Springer, Singapore

DOI: doi.org/10.1007/978-981-33-6835-4_8

Research Projects

Explaining Predictions: Brain MRI image regions responsible for tumor detection

Part one of this project is to detect cancer type(no_tumor,glioma_tumor, meningioma_tumor,pituitary_tumor) using VGG16 pre-trained model and running custom training on our Brain MRI dataset.
The second part explains the regions in the MRI image that is most important for the classification of the images
XAI or explainable AI for image data shows which inputs( pixels in images) are responsible for the prediction by the deep learning model
In this case, we have used Layerwise Relevance Propagation (LRP)
For in depth study on the implementation of LRP you can follow this:https://git.tu-berlin.de/gmontavon/lrp-tutorial

Github: [Link]

People Counting System in Public Transport

people count demo.mp4

Used YOLO v5 Object detection pretrained model for detecting and counting persons in public transports

Medicine Name Recognition from Handwritten Prescriptions

Automatic Handwritten Prescription Reader.mp4

An Optical Character Recognition (OCR) based application that can extract Medicine Names and other information from a Hand-written prescription.

End to End Encrypted Steganography Pipeline for Sensitive Data Security

Steganography : The word Steganography is derived from two Greek words- ‘stegos’ meaning ‘to cover’ and ‘grayfia’, meaning ‘writing’, thus translating to ‘covered writing’, or ‘hidden writing’. Steganography is a method of hiding secret data, by embedding it into an audio, video, image, or text file. It is one of the methods employed to protect secret or sensitive data from malicious attacks. [Source : GeeksForGeeks]

To pass sensitive images or text through a network, we have

Encrypted the sensitive image using AES encryption
A carrier image is used so that we can hide our encrypted image in it
LSB Steganography is used for hiding the encrypted image into the carrier image

Github : [Link]

Search Keyword Based Bangladeshi Newspaper Scrapper

For any search keyword, this scrapper gathers all the news based on the keyword and finds the news.
This works for all the popular Bangladeshi online newspapers
A helpful tool for text mining and natural language processing tasks
Used Selenium and Beautiful Soup

Github: [Link]

Machine Learning Based Credit Customer Classification and Analysis Dashboard

To aid data driven decision making, businesses decision makers need to know beforehand which credit customers are most likely to become defaulter. The objective of this project is to classify the credit customers of Motors so that the Credit Recovery team can take necessary actions according to customer classification.

Key Points:

Based on the credit customer’s payment pattern, predict the customer type
A live system for prediction for all current and future credit customers
Graphic visualization of predicted customer’s demography such as territory and payment information

5G Base Station Configuration at Huawei Labs (Huawei HQ, Shenzhen)

Introduction to 5G technology
Formal training on the theoretical aspects of 5G
Set up complete hardware system and integrated with the pre-developed software system
Testing the live 5G network
Experiments on network security strengths and vulnerabilities

Other Projects

Ticket Sales Assistant Facebook Messenger Bot

Bot for online ticket sales.
Intent parsing is done with Chatfuel API.
PHP is used for back-end development
Usable with Facebook messenger.
To use in messenger, search for 'Ticket-Bot'

Page updated

Google Sites

Report abuse