KT
Kazi Tanvir
  • 🏠Home
  • 👤About
  • 🚀Projects
  • 📚Research
  • 📧Contact Us
  • 🏠Home
  • 👤About
  • 🚀Projects
  • 📚Research
  • 📧Contact Us

Research & Publications

Explore my academic research and published papers in various fields of study

26
Publications
11
Conferences

Filter by

Publication Type

Research Area

26 Results Found
Book ChapterDigital Twin Approaches in Autonomous Vehicles
January 1, 2027

Digital Twin-Driven Innovation in Autonomous Vehicle Systems Through Modelling, Prototyping, and Lifecycle Optimization

Mahfjur Rahman, Mirza Asif Mahmud, Noboranjan Dey, Kazi Tanvir, Zishan Ahmed Onik, Dipta Gomes

This chapter examines how Digital Twin technology supports autonomous vehicles through real-time monitoring, predictive maintenance, simulation, and performance optimization. It discusses benefits for safety, decision-making, and lifecycle management while addressing challenges such as data synchronization, computing demands, and cybersecurity. Future directions include scalable systems, V2X integration, and AI-driven optimization methods.

Read Paper
DOI
Journal ArticleHealthcare Analytics
December 1, 2026

A hybrid optimization and tree-based learning framework for dengue diagnosis using hematological data

Kazi Tanvir, Mahfujur Rahman, Dipta Gomes

This study presents an explainable machine-learning framework for early dengue diagnosis using routine complete blood count features. A Grey Wolf Optimizer tuned an Extra Trees Classifier, achieving 95.78% accuracy. SHAP and LIME identified key hematological predictors, supporting transparent, cost-effective clinical decision support for resource-limited healthcare settings in Bangladesh and beyond.

Read Paper
PDFDOI
Journal ArticleBiomedical Signal Processing and Control
October 15, 2026

PsoriaNET: An attention-based deep convolutional neural network for automated psoriasis detection with explainable AI

Md Sayem Kabir, Kazi Tanvir, Tasnim Sultana Sintheia, Dipta Gomes, Akinul Islam Jony

This study presents PsoriaNET, an explainable deep learning framework for distinguishing psoriasis from healthy skin. The model combines a custom CNN with attention mechanisms, advanced image preprocessing, and XAI methods including LIME, Grad-CAM, and Grad-CAM++. Evaluated on 6,000 augmented images, PsoriaNET achieved 98.89% accuracy, 0.9886 precision, and a Cohen’s Kappa of 0.9745, outperforming several state-of-the-art models. The results demonstrate its potential as an accurate and interpretable tool for AI-assisted psoriasis screening.

Read Paper
DOI
Book ChapterArtificial Intelligence for Medical Data Processing
August 25, 2026

Hybrid DL & XAI for histopathology & cancer biomarkers analysis

Md Sayem Kabir, Kazi Tanvir, Tasnim Sultana Sintheia, Shanzida Zaman Shimu, Dipta Gomes, Mahfujur Rahman

This study presents an interpretable deep-learning framework for classifying Biglycan-stained breast histology images. Using a newly developed dataset, extensive preprocessing, augmentation, CNNs, transformers, and a hybrid DeiT–DenseNet169 model, the approach achieved 95.67% accuracy. LIME, Grad-CAM, and Grad-CAM++ supported transparent, clinically relevant decision-making for breast cancer diagnosis and patient management.

Read Paper
DOI
Journal ArticleArtificial Intelligence and Applications
August 14, 2026

A Novel Explainable Deep Learning Model for Early-Stage Brain Tumor Classification: Multi-Level Feature Fusion from Merged MRI Datasets

Md Sadi Al Huda, Kazi Tanvir, Zubaida Akhter, Md Shahidul Khan Pappo, Md Asraf Ali, Nasim Ahmed

This study proposes an explainable hybrid deep learning model combining Tiny-ConvNeXt and DenseNet169 for brain tumor classification from MRI images. Two open-access datasets were combined and augmented to improve model generalization. The proposed framework achieved 99.74% classification accuracy, outperforming existing methods. Ablation and statistical tests confirmed model robustness, while LIME and Grad-CAM++ provided visual explanations of tumor regions. The results support its potential for accurate and interpretable clinical decision support.

Read Paper
PDFDOI
Journal ArticleNetwork Modeling Analysis in Health Informatics and Bioinformatics
August 7, 2026

Advancing immunogenic peptide identification using explainable machine learning framework for rational vaccine design

Mahin Montasir Afif, KM Kabir, Mahfujur Rahman, Dipta Justin Gomes, Kazi Tanvir, Md Mortuza Ahmmed, Md Obaidur Rahaman, Jasim Uddin

This study develops an interpretable machine learning framework for identifying immunogenic peptides using sequence and physicochemical features. The proposed modified ν-SVM with L1 feature selection achieved strong predictive performance (AUC = 0.991, accuracy = 94.9%), while SHAP and LIME provided biological interpretation of important peptide features.

Read Paper
PDFDOI
Conference Proceedings2026 International Conference on Frontiers of Engineering and Emerging Technologies (FET)
July 17, 2026

Predicting Maternal Health Risk Levels in Rural Bangladeshi Patients Using Supervised Machine Learning

MD Shakil Mia, Sheikh Risalat Sanjan, Kazi Tanvir, Kamruddin Nur

This study develops an optimized machine learning framework for maternal health risk prediction using data from 1,014 patients in rural Bangladesh. RandomizedSearchCV was applied to five classifiers, with the tuned Random Forest achieving 87.19% accuracy, 98.2% high-risk recall, and an F1-score of 0.8728. Statistical testing, learning curves, and bootstrap confidence intervals confirmed model reliability, while blood glucose and systolic blood pressure emerged as key predictors. The framework shows strong potential for low-cost maternal risk screening in remote healthcare settings.

Read Paper
PDFDOI
Conference Proceedings2026 International Conference on Frontiers of Engineering and Emerging Technologies (FET)
July 17, 2026

ML-Based Quantification of Mental Fatigue Improvement Using Binaural Beat Stimulation

Md Foysal Bhuiyan, Md Jubairul Alam, Most Sumaia Farzana Nitu, Rubaiyat Islam, Kazi Tanvir, Kamruddin Nur

This study investigates the effects of binaural beat stimulation on mental fatigue and sleep-related behavior among university students using machine learning. Pre- and post-intervention data were analyzed with Random Forest and CatBoost models, with tuned CatBoost achieving 85.79% accuracy and an F1-score of 85.74%. The findings suggest that binaural beats combined with machine learning may offer a practical approach for monitoring and reducing mental fatigue in students.

Read Paper
PDFDOI
Conference Proceedings2026 International Conference on Frontiers of Engineering and Emerging Technologies (FET)
July 17, 2026

A GraphGRU-Based Framework for Urban Traffic Congestion Prediction

Mohammad Towhidul Islam, Emrul Hassan, Zihadul Islam, Kazi Tanvir, Kamruddin Nur

This study proposes a Graph Convolutional Gated Recurrent Unit (GC-GRU) model for short-term traffic forecasting by jointly capturing spatial relationships between traffic sensors and temporal patterns. A mask-based loss function improves robustness to missing and noisy data. Evaluated on PEMS04, PEMS08, and PEMS-BAY, the model achieved competitive performance, including an average MAE of 0.25 on PEMS04, demonstrating its potential for reliable real-time traffic prediction.

Read Paper
PDFDOI
Journal ArticleInternational Journal of Computational Intelligence Systems
June 28, 2026

A Deep Learning Based Three-Stage Fusion Neural Network with Composite Adaptive Momentum Optimizer for Bangla Cyberbully Identification

Mahin Montasir Afif, Abdullah Al Noman, Kazi Abdullah Jarif, Md Emamul Arefin, Sathya Narayana Sharma K Islam, Mahfujur Rahman, Dipta Gomes, Kazi Tanvir, Md Obaidur Rahaman, Syed Mohammed Shamsul Islam

This study presents TriB-FNN, an explainable three-stage fusion neural network for detecting cyberbullying in Bangla social media text. The model integrates CNN, Bi-LSTM, and GRU architectures with a Composite Adaptive Momentum optimizer and is evaluated on three public Bangla datasets. Extensive preprocessing and data augmentation improved generalization in the low-resource setting. TriB-FNN achieved 99.06% accuracy, with precision, recall, and F1-score above 99%, outperforming baseline models. LIME was also applied to provide interpretable predictions, demonstrating the model’s potential for reliable Bangla cyberbullying detection.

Read Paper
PDFDOI
Journal ArticleMalaysian Journal of Science and Advanced Technology
June 24, 2026

Enhancing Dengue Outbreak Prediction in Bangladesh: A Weighted Average Ensemble Machine Learning Approach

Mahfujur Rahman, Noboranjan Dey, Nazmus Sakib, Rukaiya Jahan Sajuti, Mehedi Hasan, Abdullah Hel Azmain, Rahul Biswas, Dipta Justin Gomes, Kazi Tanvir, Mirza Asif Mahmud

This study proposes a Weighted Average Ensemble Learning (WAEL) model for dengue prediction in Bangladesh by combining SVM, Random Forest, and AdaBoost classifiers. Using 199 survey records with nine attributes, the proposed model achieved 93% accuracy, 94% precision, 98.5% recall, and an F1-score of 0.955, outperforming individual classifiers. The results demonstrate the potential of ensemble learning for reliable dengue risk prediction and data-driven public health planning.

Read Paper
PDFDOI
Conference Proceedings2026 IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence & Networking (QPAIN)
June 11, 2026

Automated Tea Leaf Disease Classification Using ViT-Resnet with Explainable AI

Md Sayem Kabir, Mst Mafroza Haque Iqra, Laboni Somoddar, Kazi Tanvir, Farzana Nazera, Dipta Gomes

This study proposes ViTResNet, a hybrid deep learning framework combining Vision Transformer and ResNetV2 for tea leaf disease classification. Evaluated on a balanced dataset of 12,000 images across four classes, the model achieved 99.83% accuracy, an MCC of 0.992, and Cohen’s Kappa of 0.995. Comparative and ablation studies confirmed the effectiveness of combining global and local visual features. The framework offers a highly accurate approach for automated tea disease detection and timely crop management.

Read Paper
PDFDOI
Conference Proceedings2026 IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence & Networking (QPAIN)
June 11, 2026

An Interpretable Hybrid Ensemble Framework for Medium-Term Electricity Demand Forecasting in Urban Power Systems

Samiha Tasnim, Tousif Tarik, Kazi Tanvir, Kamruddin Nur

This study proposes a hybrid medium-term power demand forecasting framework for Dhaka City by integrating LightGBM, XGBoost, CatBoost, and a Deep Neural Network. Evaluated on daily electricity consumption and temperature data from 2021–2025, the model achieved an RMSE of 139.535 MW, MAE of 111.1501 MW, and \(R^2\) of 0.9530. SHAP and LIME provided interpretable forecasts, while an auxiliary load-state classifier achieved 96.44% accuracy. The framework offers a reliable approach for urban energy planning and smart-grid management.

Read Paper
PDFDOI
Conference Proceedings2026 IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence & Networking (QPAIN)
June 11, 2026

An Explainable Cat-Boost Framework for Predicting Cesarean Section Delivery Utilization Patterns in Bangladesh

Mohammad Tofiqul Islam, Hasan Mahmud Shanto, Supprio Saha Pranto, Kazi Tanvir, Kamruddin Nur

This study develops a machine learning framework to predict Cesarean section delivery among Bangladeshi women using 2,442 records from the BDHS 2022 dataset. Multiple models were evaluated, with CatBoost achieving the best performance at 88.55% accuracy, 0.9515 ROC-AUC, and an F1-score of 0.8852. Maternal education, wealth index, and antenatal care visits emerged as important predictors. The findings can support data-driven strategies to identify and reduce potentially unnecessary Cesarean deliveries in Bangladesh.

Read Paper
PDFDOI
Conference Proceedings2026 IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence & Networking (QPAIN)
June 11, 2026

GDR-GNN: A Graph Neural Network Framework for Explainable Hereditary Genetic Disorder Risk Prediction

Arpon Paul Amit, Saiful Islam Oni, Kazi Tanvir, Dipta Justin Gomes, Mahfujur Rahman, Muhibul Haque Bhuyan

This study proposes GDR-GNN, a graph neural network for predicting hereditary genetic disorder risk from clinical and genetic data represented as a k-nearest-neighbor graph. Optuna was used for hyperparameter optimization, while SHAP and LIME provided model interpretability. Evaluated on data from 100 families, GDR-GNN achieved 97.67% accuracy and a Cohen’s Kappa of 0.9651, outperforming conventional machine learning models. The results demonstrate the potential of explainable graph-based learning for reliable hereditary risk prediction.

Read Paper
PDFDOI
Conference Proceedings2026 IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence & Networking (QPAIN)
June 11, 2026

Noise-Robust Speech Emotion Recognition Using Deep Learning: Towards Real-Life Adaptive Emotion-Aware Systems

Bijoy Ahamed, Kazi DKM Imtiaz, Riaz Raihan Niloy, Kazi Tanvir, Kamruddin Nur

This study proposes a reinforcement learning-based adaptive noise augmentation framework to improve Speech Emotion Recognition in noisy environments. Unlike fixed or random augmentation, the method dynamically selects the type, intensity, and position of added noise during training. Experiments on RAVDESS with NOISEX-92 using CNN, CNN-BLSTM, and Transformer models achieved a 7–9% absolute accuracy improvement over conventional noise augmentation. The framework demonstrates strong potential for robust real-world speech emotion recognition.

Read Paper
PDFDOI
Conference Proceedings2026 IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence & Networking (QPAIN)
June 11, 2026

Uncertainty-Aware and Explainable Feature-Level Ensembles for CT-Based Kidney Stone Detection

Abdullah Khondoker, Faria Ahmed Richi, Sarah Binte Islam, Kazi Tanvir, Dipta Justin Gomes, Mahfujur Rahman

This study proposes a feature-level ResNet18–DenseNet121 ensemble for CT-based kidney stone classification. Using CT-specific preprocessing and augmentation, the model achieved 99.44% accuracy, an F1-score of 0.9944, and ROC-AUC of 0.9993, outperforming individual backbones. Cross-validation confirmed robust generalization, while Monte Carlo dropout, LIME, SHAP, and Grad-CAM provided uncertainty estimation and interpretable predictions. The framework shows strong potential for reliable and explainable kidney stone screening from CT images.

Read Paper
PDFDOI
Journal ArticleSN Computer Science
June 8, 2026

Husk Species Classification Using a ViT–DenseNet Hybrid Model with Explainable AI

Dipta Gomes, Kazi Tanvir, Md Sayem Kabir, Tasnim Sultana Sintheia, Sadman Samir Rafith, Mohammad Ariyan Pathan

This study proposes a hybrid ViT–DenseNet121 model for automated classification of eight agricultural husk species using the BDHusk dataset. The model achieved 97.23% test accuracy and an MCC of 0.9864, supported by preprocessing, feature fusion, and transfer learning. LIME, Grad-CAM, and Grad-CAM++ were used to provide interpretable predictions. The framework shows strong potential for automated sorting and sustainable agro-industrial applications.

Read Paper
PDFDOI
Journal ArticleSN Computer Science
June 4, 2026

ST-HRF: A Swin Transformer–Harmony Optimised Random Forest Framework for Poultry Health Sound Analysis

Md Sayem Kabir, Kazi Tanvir, Sadman Samir Rafith, Mohammad Ariyan Pathan, Md Sadi Al Huda, Touhid Bhuiyan

This study proposes ST-HRF, a hybrid framework for automated poultry health classification using chicken vocalisations. Mel-spectrograms, acoustic features, Swin Transformer embeddings, and a Harmony Search-optimised Random Forest were combined for prediction. The model achieved 98.12% accuracy, MCC of 0.9784, and Cohen’s Kappa of 0.9596, outperforming baseline models. SHAP and attention maps improved interpretability, supporting scalable and non-invasive poultry disease monitoring.

Read Paper
PDFDOI
Journal ArticleInformatics and Health
April 2, 2026

An explainable ensemble learning framework for ovarian cancer classification using blood biomarkers

Kazi Tanvir, Md Sayem Kabir, Kazi Tasnim Hena, Tasnim Sultana Sintheia, Shanzida Zaman Shimu, Dipta Gomes, Mahfujur Rahman, Mirza Asif Mahmud

This study proposes an interpretable machine learning framework for distinguishing ovarian cancer from benign ovarian tumors using routine blood biomarkers. A soft-voting ensemble of Histogram-Based Gradient Boosting and K-Nearest Neighbors, combined with ROSE resampling and LASSO feature selection, achieved 98.61% accuracy and F1-score. Multiple XAI methods, including SHAP, LIME, and surrogate decision trees, provided transparent predictions. The framework shows strong potential for accurate and interpretable ovarian cancer screening.

Read Paper
PDFDOI
Book ChapterVision Language Models for Next-Generation Healthcare
April 1, 2026

Vision Language Models in Healthcare Through a Multimodal Approach to Medical Imaging and Clinical Applications

Kazi Tanvir, Rahul Biswas, Dipta Gomes, Mahfujur Rahman, Noboranjan Dey, Md Reazul Islam

This chapter explores the emerging role of Vision Language Models (VLMs) in healthcare, focusing on their ability to integrate visual and textual data to improve medical imaging analysis and clinical decision-making. It examines the key components of VLMs, including vision and language models, their fusion techniques, and their applications in tasks like disease detection, report generation, and visual question answering. The chapter also addresses challenges such as data scarcity, privacy concerns, and model interpretability. Additionally, it highlights future directions for enhancing model generalization, improving explainability, and ensuring seamless integration into clinical workflows, while emphasizing the importance of ethical considerations and real-world validation for safe deployment in healthcare environments.

Read Paper
PDFDOI
Conference ProceedingsProceedings of Fifth International Conference on Computing and Communication Networks
April 1, 2026

ViTBiT-PoxNet: An Explainable Hybrid Deep Learning Framework for Enhanced Early-Stage Monkeypox Classification

Md Sadi Al Huda, Tahmid Enam Shrestha, Kazi Tanvir, Rafat Ahmed, Md Asraf Ali, Touhid Bhuiyan

This study proposes a hybrid deep learning model combining ViT-B/16 and BiT-M-R50x1 for automated monkeypox classification from skin images. Trained on 3,192 images, the model achieved 99.07% accuracy, 99% precision, 98.87% recall, and an F1-score of 99.16%. LIME and Grad-CAM++ were used to provide interpretable predictions. The framework shows strong potential for accurate and explainable monkeypox screening.

Read Paper
PDFDOI
Journal ArticleData in Brief
February 24, 2026

SSC-BanglaTutor: A curriculum-aligned Bengali dataset for intelligent tutoring systems

Eshraque Jabid Ifti, Fihab Ifty, Mehadi Hasan, Rahul Chandra Shil, Utshab Kumar Saha, Kazi Tanvir, Mahfujur Rahman, Dipta Gomes

This dataset provides 11,286 Bengali hint-based question-answer entries for SSC-level Biology, Chemistry, and Physics in Bangladesh. Developed from government textbooks, study materials, and past exam questions, each entry includes a correct answer, related distractors, and a convergence score indicating the expected hint progression. The dataset supports personalized learning, learner modeling, and fine-tuning of LLM-based intelligent tutoring systems for low-resource Bengali educational applications.

Read Paper
PDFDOI
Conference ProceedingsProceedings of the 3rd International Conference on Big Data, IoT and Machine Learning
February 20, 2026

Transformer-Based and Probabilistic Approaches for Topic Modeling in News Article Analysis

Md Bakibillah Rahat, Tasnim Sultana Sintheia, Kazi Tanvir

This study presents a transformer-based topic modeling framework for extracting latent themes from large-scale news articles collected from *The Daily Star*. Using BERTopic and BGE-Base-en-1.5 embeddings, the framework was evaluated through coherence, diversity, purity, and topic distribution metrics. BGE-Base-en-1.5 achieved the highest coherence of 0.69, while BERTopic obtained the best cluster purity of 0.95. The results demonstrate the effectiveness of embedding-based topic modeling for scalable and interpretable news analysis.

Read Paper
PDFDOI
Journal ArticleAdvances in Biomarker Sciences and Technology
January 14, 2026

FRF-HHO: Early ovarian cancer prediction using explainable fuzzy random forest optimized by Harris Hawks algorithm

KM Yeaser Arafat, Ahmed Hossain, Mushfika Ikfat, Md Areful Amin, Kazi Tanvir, Dipta Gomes, Mahfujur Rahman

This study proposes an interpretable ovarian cancer prediction framework combining Fuzzy Random Forest with Harris Hawks Optimization using clinical and biochemical data from 349 patients. After RFECV feature selection and SMOTE–Tomek balancing, the optimized model achieved 94.12% accuracy, 96.07% recall, and an F1-score of 93.69%. SHAP and LIME identified AFP, HE4, CA125, and age as important predictors. The results show promising potential for early ovarian cancer decision support, although larger multi-center validation is needed.

Read Paper
PDFDOI
Book ChapterAgentic AI for Autonomous Vehicles: Safety, Reliability, Law, and Ethics
January 1, 2026

A Comprehensive Study on the Ethical Perspective of Integration of Agentic AI for Autonomous Vehicles

Mirza Asif Mahmud, Mahfujur Rahman, Dipta Gomes, Kazi Tanvir, Md Reazul Islam

This chapter explores the ethical integration of Agentic AI into autonomous vehicles, detailing a technical architecture that embeds ethical considerations across goal setting, memory, human oversight, transparency, and robustness. It highlights how V2X connectivity transforms AVs into collaborative agents within intelligent ecosystems. The chapter also presents evaluation frameworks and metrics to measure safety, fairness, and public trust, ensuring autonomous driving aligns with societal values and legal standards.

Read Paper
PDFDOI

Interested in Collaboration?

I'm always open to discussing research opportunities and collaborations.

Get in Touch
KT

Kazi Tanvir

Data Scientist & Machine Learning Engineer specializing in AI research, deep learning, and neural networks. Building innovative solutions through code and research.

📧kazitanvir.bangladesh@gmail.com

Quick Links

  • →Home
  • →About
  • →Projects
  • →Research
  • →Contact

Stay Updated

Subscribe to get latest updates and insights.

© 2026 Kazi Tanvir. All rights reserved.

Developed by Md Rijoan Maruf

Follow: