ongoing works


My Undergraduate thesis Based Project

Explainable AI (XAI) Powered Transparent Network Intrusion Detection System to Enhance Cybersecurity

This study focuses on leveraging Machine Learning (ML) and Deep Learning (DL) techniques to create innovative frameworks that incorporate eXplainable Artificial Intelligence (XAI) to enhance intrusion detection and analysis across various network environments. The primary aim is to address the "black-box" nature of current Deep Learning and Machine Learning-based Intrusion Detection Systems (IDSs) by elucidating the decision-making process behind these systems. To achieve this objective, we have developed eight distinct ML/DL architectures, based on Decision Trees (DT), Ensemble Trees (ET), eXtreme Gradient Boosting (XGB), Random Forest with Recursive Feature Elimination (RFC+RFE), Multi-Layer Perceptrons (MLP), Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), and Autoencoders. These architectures incorporate XAI techniques such as Local Interpretable Model-agnostic Explanations (LIME), SHapley Additive exPlanations (SHAP), ELI5, and ProtoDash to provide clear rationales for critical decisions made by the IDSs. The overarching goal is to enhance the transparency of IDSs, enabling cybersecurity professionals and human analysts to comprehensively understand their functioning. This research involves the creation of multiple IDSs using both traditional machine learning and advanced deep neural networks designed to swiftly detect and anticipate network intrusions. Subsequently, XAI techniques are integrated into these architectures to offer explanations for model classifications, the reasoning behind predictions, and the importance of features in prediction accuracy. A comprehensive evaluation is conducted, including extensive testing with well-known attack scenarios, to assess the effectiveness and interpretability of our proposed framework. The results of these tests demonstrate the efficiency and clarity of our methodology, highlighting its precision in identifying and deducing various threats. This study provides valuable insights into the benefits of combining ML/DL approaches with XAI in the context of network security, enhancing the reliability, transparency, and justification of the outcomes.

INDEX TERMS Intrusion detection system, cybersecurity, cyber threat intelligence, explainable AI, machine learning, deep learning, LIME, SHAP, ELI5, ProtoDash, local and global explanations.

[ I will be more than happy to share my project paper and extensive coding activities carried out for this project with potential grad school supervisors ]

Published and under review works

Peer-Reviewed Journal Publications

Md. Tohidul Islam, Md. Khalid Syfullah, Md. Golam Rashed, Dipankar Das. Bridging the Gap: Advancing the Transparency and Trustworthiness of Network Intrusion Detection with Explainable AI. Accepted, International Journal of Machine Learning and Cybernetics. PREPRINT available at Research Square [https://doi.org/10.21203/rs.3.rs-3263546/v1]

With the explosive rise of internet usage and the development of web applications across various platforms, ensuring network and system security has become a critical concern. Networks and web services are particularly susceptible to targeted attacks, as hackers and intruders persistently attempt to gain unauthorized access. The integration of artificial intelligence (AI) has emerged as a crucial tool for detecting intrusions and constructing effective Intrusion Detection Systems (IDSs) to counter cyber-attacks and malicious activities. IDSs developed using machine learning (ML) and deep learning (DL) techniques have proven to be highly effective in detecting network attacks, offering machine-centric solutions. Nevertheless, mainstream adoption, confidence and trust in these systems have been greatly impeded by the fact that ML/DL implementations tend to be “black boxes,” and thus lacking human interpretability, transparency, explainability, and logical reasoning in their prediction outputs. This limitation has prompted questions about the responsibility and comprehension of AI-driven intrusion detection systems. In this study, we propose four novel architectures that incorporate Explainable Artificial Intelligence (XAI) techniques to overcome the challenges of limited interpretability in ML/DL based IDSs. We focus on the development of ExplainDTC, SecureForest-RFE, RationaleNet, and CNNShield architectures in network security solutions, and inquiry into their potential to convert the untrustworthy architectures into trustworthy. The models are applied to scan network traffic and identify, and report intrusions based on the traits extracted from the UNSW-NB15 dataset. To explain how a decision is made by the models and to add expansibility at every stage of machine learning pipeline, we integrate multiple XAI methods such as LIME, SHAP, ELI5, and ProtoDash on top of our architectures. The generated explanations provide quantifiable insights into the influential factors and their respective impact on network intrusion predictions.

Keywords: Intrusion detection system, cybersecurity, explainable AI, machine learning, deep learning, LIME, SHAP, ELI5, ProtoDash, local and global explanations.

MFA, MKS, OS, MTI, MN, MRI, AK, MAA, MEHC. IRv2-Net: A Deep Learning Framework for Enhanced Polyp Segmentation Performance Integrating InceptionResNetV2 and UNet Architecture with Test Time Augmentation Techniques.  "Sensors Journal."  Sensors 2023, 23, 7724. — IF: 3.9 — https://doi.org/10.3390/s23187724

Colorectal polyps in the colon or rectum are precancerous growths that can lead to a more serious disease called colorectal cancer. Accurate segmentation of polyps using medical imaging data is important for effective diagnosis. However, manual segmentation by endoscopists can be time-consuming, error-prone, and expensive, leading to a high rate of missed anomalies. To solve this problem, an automated diagnostic system based on deep learning algorithms is proposed to find polyps. The proposed IRv2-Net model is developed using the UNet architecture with a pre-train InceptionResNetV2 encoder to extract the most features from the input samples. The Test Time Augmentation (TTA) technique, which utilizes the characteristics of the original, horizontal, and vertical flips, is used to gain precise boundary information and multi-scale image features. The performance of numerous state-of-the-art (SOTA) models are compared using several metrics such as accuracy, Dice Similarity Coefficients (DSC), Intersection Over Union (IoU), precision, and recall. The proposed model is tested on the Kvasir-SEG and CVC-ClinicDB datasets, demonstrating superior performance in handling unseen real-time data. It achieves the highest area coverage in the area under the Receiver Operating Characteristic (ROC-AUC) and area under Precision-Recall (AUC-PR) curves. The model exhibits excellent qualitative testing outcomes across different types of polyps, including larger, smaller, over-saturated, sessile, or flat polyps, within the same dataset and across different datasets. Through our approach, the number of missed rating difficulties can be significantly minimized. Lastly, we create a graphical interface for producing the mask in real-time. The findings of this study have potential applications in clinical colonoscopy procedure and can serve on the basis of further research and development.

Keyword: Segmentation, Colonoscopy, IRv2-Net, Kvasir-SEG, CVC-ClinicDB, Test-Time Augmentation, Polyps.

Contribution: Methodology, formal analysis, investigations, coding, and writing.

O. Sarkar, M. R. Islam, M. K. Syfullah, M. T. Islam, M. F. Ahamed, M. Ahsan, and J. Haider. Multi-Scale CNN: An Explainable AI Integrated Unique Deep Learning Framework for Lung Affected Disease Classification. Technologies, MDPI. Technologies 2023, 11, 134.

The lung-affected diseases are still the leading cause of human death worldwide. To make diagnostics decisions faster, it is highly important to detect the diseases quickly and accurately. However, it is challenging to make available a significant number of testing apparatus, and they also have a lower level of reliability. Recent research indi-cates the usefulness of Chest X-Ray images in identifying lung diseases for instance COVID, fibrosis, pneumonia, and many others. In this study, four publicly available datasets were combined to produce a dataset of 6650 CXR images, which were divided into seven disease groups. To distinguish between normal and six lung-related illness-es (bacterial pneumonia, COVID, fibrosis, lung opacity, tuberculosis, and viral pneu-monia), a Deep Learning (DL) architecture called Multi-Scale CNN was devised that can merge predictions from multiple feature maps at different resolution scales to im-prove class predictions. To assess the model's effectiveness, accuracy, precision, recall, f1-score, and area under the curve (AUC) were utilized. With an accuracy of 96.05% and average values for precision, recall, f1-score, and AUC of 0.97, 0.95, 0.95, and 0.94, the results showed that the suggested model surpassed other popular transfer learning (TL) models such as VGG16 and VGG19 for a seven-class classification. The CNN model also performed well when other multiclass classifications with accuracy values for two, three, four, five, and six classes, respectively, were 100%, 99.65%, 99.21%, 98.67%, and 97.47%. The proposed model also outperformed all previous state-of-the-art (SOTA) works applied to multi-class CXR data to diagnose lung-affected diseases.

Keywords: COVID; Chest X-Ray (CXR) image; Deep Learning; Multi-Scale CNN; Feature map,

Hossen, I., Islam, M.T., Rashed, M.G., Das, D., S.M.A.H.M ALIM, (2023). Suicide Prevention: A Text-based AI system for Predicting Depression, Anxiety, and Stress (DAS) Levels of Bengali Speaking Facebook Users. Under Review in ACM Transactions on ”Asian and Low-Resource Language Information Processing journal."

Facebook is undoubtedly a good mood of communication tool to be socially connected and updated. Facebook is being used for sharing personal thoughts including mental health. Even suicidal ideation is being shared. Many people can see these, but rarely get success in preventing suicide. However, technology could be used to mitigate suicidal ideation or even prevent suicide. Depression, anxiety, and stress are the key factors in deciding suicide. There is no previous work in Bangla that considered Depression, Anxiety, and Stress (DAS) for suicide prevention. Therefore a text-based Artificial Intelligence system was proposed for Predicting DAS levels that can potentially help to prevent suicide. The Bengali Facebook Emotion Depression Anxiety Stress (BenFEDAS) dataset was introduced A DASNet and DepNet two neural network models were introduced. Finally, an Anti-suicidal resource recommendation(ASR2) system was proposed for suicide prevention. Seven popular machine learning (ML) algorithms were used; DASNet (F1-score of 52%) and DepNet (F1-score of 71%) for validating the acceptability of our proposed model. The result was compared with other standard approaches. The result showed that DASNet and DepNet outperformed other ML algorithms. Furthermore, the accuracy of the DASNet and DepNet is comparable with other standard methods.

Additional Key Words and Phrases: Social Media, Facebook, Depression, Emotion, Suicide Prevention, Classification, Logistic Regression, Support Vector Machine, Recurrent Neural Network, FastText

Contribution: Lead author and I contributed equally to this work from conceptualization to development.

Peer-Reviewed Book Chapter Publication

Hossen, I., Islam, T., Rashed, M.G., Das, D. (2022). Early Suicide Prevention: Depression Level Prediction Using Machine Learning and Deep Learning Techniques for Bangladeshi Facebook Users. Lecture Notes in Networks and Systems, vol 437. Springer, Singapore.

Depression is a crucial factor for deciding to suicide. However, few works have been done on depression analysis using the Bengali language based on social media data such as Facebook. In this paper, we propose a depression detection model for Facebook users using Logistic Regression and LSTM. This work aims to analyze the status updates from Facebook users within 2–3 years and evaluate them to detect whether the person is depressed or not. We collected data from 100 users’ profiles from Facebook, containing on average 30 posts from each user, and proposed BenFED dataset. The proposed system considers sixteen emotional factors related to depression. Based on these emotions, Facebook users’ statuses are labeled as four types, e.g., no, mild, moderate, and severe for determining the level of depression. We compared our proposed approach with other state-of-the-art approaches. It is revealed that our proposed approach outperformed most of the compared techniques for detecting emotions, depression levels, and depression statuses.

INDEX TERMS  Social media - Facebook - Depression - Emotion - Suicide prevention - Classification - Logistic regression - LSTM

Contribution: Lead author and I contributed equally to this work from conceptualization to development.

Peer-Reviewed International Conference Publications

O. Sarkar, M. R. Islam, T. Hossain, M. K. Syfullah, M. T. Islam, and M. Moniruzzaman, An Empirical Model of Classifying Lung Affected Diseases to Detect COVID-19 Using Chest X-ray Employing Convolutional Neural Architecture, 2022 IEEE International Conference on Artificial Intelligence in Engineering and Technology (IICAIET), Kota Kinabalu, Malaysia, 2022.

The earlier detection and accurate diagnosis of COVID seem to be a global problem. It is difficult to make a large number of testing equipment, but then again, their reliability is relatively poor. Recent research indicates the usefulness of chest x-ray pictures in identifying COVID. This study presents a deep learning algorithm developed from the ground up to categorize as well as confirm the existence of COVID in a set of X-ray imaging data. We designed a CNN architecture from the ground up to retrieve elements from provided X-ray data to categorize them and identify the individual contaminated with COVID. Our strategy may aid in mitigating the consistency issues while working with medical data. In contrast to some other classifying activities with a large enough image database, obtaining large X-ray datasets for this classification job is challenging. So, we applied multiple data enhancement techniques to maximize the accurateness, achieving a significant accuracy of 97.75 percent.

Index Terms— COVID, RELU, Softmax, AUC-ROC, Grad- CAM.

Contribution:  Conceptualization, Literature review, Methodology, Experimental setup and coding, writing.


M. T. Islam, M. K. Syfullah, J. Islam,H. M. S. Quadir, M.G. Rashed, D. Das. Exploring with Explainable AI: Machine Learning vs. Deep Learning in Network Security. 26th International Conference on Computer and Information Technology, IEEE. IEEE Xplore, doi: 10.1109/ICCIT60459.2023.10441363.


The increasing complexity and diversity of cyber threats have driven the widespread adoption of Machine Learning (ML) and Deep Learning (DL) techniques for Network Intrusion Detection Systems (NIDSs) due to their capacity to effectively adapt and detect evolving threats. However, the inherent black-box nature of ML and DL-powered NIDSs has raised concerns surrounding explainability, transparency, and interpretability. This challenge has led to a prevailing sense of distrust in these systems. In this paper, we propose an improved approach to enhance the intrusion detection capabilities of Network Intrusion Detection System (NIDS) by introducing a ML and a DL-based classifier. To overcome the explainibility and trust issues, we innovatively integrate advanced Explainable Artificial Intelligence (XAI) techniques, such as SHAP, LIME, and ELI5, with our proposed ML and DL classifiers. These techniques provide vital insights and detailed explanations behind the decision-making processes of the NIDS. This approach is pivotal for multiple reasons, including ensuring that the classifiers are fair and unbiased, articulating the rationale behind their outcomes, or for debugging the models when they make mistakes. Furthermore, we evaluate and compare the performance of ML and DL classifiers using a variety of metrics such as Accuracy, Precision, Recall, F1-score, ROC AUC curve, PR AUC, Training time, Confusion matrix, and also compare the explainability the classifiers. The need to compare ML and DL based classifiers stems from the aim to optimize network security measures, improve threat detection, ensure that resources are used efficiently in the face of evolving cyber threats, and understanding the differences in their decision-making processes in network intrusion detection.


Keywords—Network Security, Intrusion Detection Systems, ML, DL, Explainable AI, Cybersecurity


J. Islam, M. T. Islam, M.G. Rashed, D. Das. Accurate Vehicles Detection and Speed Estimation Using Homography Based Background Subtraction and Deep Learning Approaches. 26th International Conference on Computer and Information Technology, IEEE. IEEE Xplore, doi: 10.1109/ICCIT60459.2023.10441114.


In the context of the ever-evolving landscape of information and communication technology, urban populations worldwide are increasingly embracing the notion of smart cities. Smart transportation, a fundamental component of the smart cities, falls under the purview of what is commonly referred to as the Intelligent Transportation System (ITS). This system plays a pivotal role in the management of highway transportation infrastructure. A key facet of this system involves the widespread installation of Closed Circuit Television (CCTV) cameras on urban thoroughfares. These cameras serve a dual function, dili- gently monitoring traffic conditions and detecting anomalies such as traffic congestion and violations of prescribed vehicle speed limits. This research study is primarily dedicated to the task of estimating vehicle speeds using two distinct methodologies: background subtraction based approach and deep learning based approach. Both methodologies leverage the concept of inverse perspective projection to achieve precise vehicle detection and speed estimation. The study is underpinned by the utilization of two distinct datasets, one for the purpose of training and the other for estimating vehicle speeds. The initial phase of our investigation focuses on the accurate detection of vehicles. To accomplish this, we trained a YOLOv8 model, yielding impressive outcomes. In the realm of vehicle detection accuracy, the background subtraction based method achieved an accuracy rate of 92.14%, while the deep learning approach demonstrated an even higher level of accuracy, standing at around 98.88%. Then, our research shifts its focus to the tracking of vehicles frame-by-frame and the subsequent calculation of their speeds. This is achieved with the aid of two reference lines in both the both vehicle detection methodologies. The results obtained from our experiments unequivocally highlight the superior speed esti- mation capabilities of deep learning approaches when compared to the background subtraction based method.