Asif Newaz - Research

My research focuses on developing trustworthy, data-efficient, and interpretable machine learning systems that can generalize reliably under real-world conditions —where data are often scarce, noisy, or imbalanced. I aim to design data-centric algorithms that adapt learning to sample difficulty, enhance fairness, and ensure transparency in decision-making.

I work across multiple domains, including computer vision, biomedical signal and image analysis, spatiotemporal modeling, and sensing applications, where robust and explainable AI can drive impactful solutions. I have developed open-source Python libraries such as iCost and iBRF, extending scikit-learn with data–complexity–based learning and imbalanced data solutions.

Computer Vision & Biomedical AI

I work on developing deep learning frameworks for medical imaging and visual recognition tasks, focusing on accessibility, interpretability, and reliability. My works integrate explainable AI (XAI) techniques to enhance trust in AI-assisted diagnosis.

Representative Works:

An End-to-End Deep Learning Framework for Arsenicosis Diagnosis Using Mobile-Captured Skin Images – Arxiv
– Developed a mobile-based diagnostic system for arsenicosis detection using Swin Transformer and Grad-CAM/LIME visualization. The goal is to provide an early, accessible diagnostic framework based on mobile-captured images, instead of dermoscopic images, for the rural and underprivileged population.
A data-driven Biomarker Identification Study for Ovarian Cancer Using – published in Informatics in Medicine Unlocked (paper)
– Integrated explainable AI (SHAP) models to assess the reliability of prediction frameworks and identified biomarkers from clinical datasets.

Automatic Recognition and Translation of Bangla Braille Texts Using YOLO – Work in progress
– Designing a real-time OCR pipeline for Braille character recognition and translation from images, accompanied by the curation of a novel dataset comprising Braille book and text images.
An intelligent decision support system for the accurate diagnosis of cervical cancer - published in Knowledge-Based Systems (paper)

– Developed a machine-learning-based decision-support system to diagnose cervical cancer using clinical risk-factor data (EHR analysis).

Data-Efficient and Imbalanced Learning

My core research addresses the challenges of learning from imbalanced, scarce, and noisy data. I explore instance-complexity–based frameworks, cost-sensitive algorithms, and hybrid sampling techniques to improve generalization.

Representative Works:

M.Sc. Thesis: Exploring New Frontiers in Imbalanced Learning: Data Complexity-Based Solutions (Link)
– Proposed three novel algorithms leveraging data complexity to guide sampling and cost-sensitive classification. Details can be found here.
iCost – A novel instance-complexity–based cost-sensitive learning framework to obtain better generalization than the traditional cost-sensitive approach. (Arxiv)
iBRF – An improved version of the traditional Balanced Random Forest (BRF) classifier for better, well-generalized performance under extreme imbalance. (paper)
Data Fidelity Toolkit: A SHAP-like framework to assess the stability and reliability of synthetic data generated by sampling or generative models. (in development)

Spatiotemporal Modeling & Smart Systems

I apply ML techniques to spatiotemporal and environmental datasets for predictive modeling, mobility optimization, and energy analytics.

Representative Works:

City-Scale Taxi Dispatch Optimization for New York City – Designed a dynamic dispatch optimization model integrating real-time demand forecasting. (Work in progress)
Bangladesh Power Demand Dataset (2015–2022) – Data in Brief, 2024 (Link)
– Curated a multi-year electricity demand, generation, load shedding, and external conditions dataset in Bangladesh for ML-based load forecasting research.

Trustworthy & Explainable AI

Transparency and interpretability are central to my research philosophy. I design explainable, robust, and fair AI systems that allow practitioners to understand and trust model predictions.

Representative Contributions:

Integration of SHAP, Grad-CAM, and LIME to enhance interpretability in diagnostic models.
Development of complexity-aware learning pipelines that detect unreliable or biased samples.
Ongoing research: Developing a SHAP-like toolkit for fidelity and stability assessment of augmented samples.

Open-Source Tools

iCost: Python package implementing Instance-complexity–based cost-sensitive learning framework — PyPI | GitHub
iBRF: Python package implementing Improved Balanced Random Forest Classifier — PyPI | GitHub

Datasets & Resources

Energy Consumption: Multi-year dataset on daily electricity demand, generation, load shedding, and external conditions in Bangladesh - Mendeley Data
Mobile-Captured Skin Lesion Dataset - Curated a dataset of mobile-captured skin lesion images for developing lightweight, accessible diagnostic systems for dermatological conditions. The dataset supports research on data-efficient and explainable deep learning for real-world healthcare applications. - GitHub

Collaborations

I actively engage in collaborative research across multiple domains, from wearable sensors to cybersecurity, focusing on the application of machine learning and data science in these fields.

Collaboration with KFUPM: Co-research with the team at King Fahd University of Petroleum and Minerals (KFUPM) on the use of deep learning techniques in multimode-fiber specklegram sensors in different sensing applications. (paper published in IEEE Sensors Journal)
Cybersecurity AI: Log Anomaly Detection via ML/DL/GANs - Joint work with researchers from Western University (Australia) on leveraging unsupervised machine learning techniques for detecting log-based anomalies in cybersecurity systems. (on-going research)

Mentoring

I regularly supervise undergraduate students on their final year theses and voluntary research projects. Some notable examples include -

Survival Analysis of Breast Cancer Patients: A Population-Based Study from SEER (paper)
Brain tumor detection using CNNs from MRI data (paper)
Alzheimer’s Disease Prediction from MRI data using CNNs (paper)
Leveraging Machine Learning in Material Science: Application in Perovskite Materials.

MSc Thesis

Details of my MSc thesis can be found here.

Recently Completed Projects

iCost: A Novel Instance Complexity-Based Cost-Sensitive Learning Framework

Background: Class imbalance in data presents significant challenges for classification tasks. It is fairly common and requires careful handling to obtain desirable performance. Traditional classification algorithms become biased toward the majority class. One way to alleviate the scenario is to make the classifiers cost-sensitive. This is achieved by assigning a higher misclassification cost to minority-class instances.

Research Gap: One issue with this implementation is that all the minority-class instances are treated equally, and assigned with the same penalty value. However, the learning difficulties of all the instances are not the same. Instances that are located in the overlapping region or near the decision boundary are harder to classify, whereas those further away are easier. Without taking into consideration the instance complexity and naively weighting all the minority-class samples uniformly, results in an unwarranted bias and consequently, a higher number of misclassifications of the majority-class instances.

Proposal: To alleviate the situation, we propose a novel instance complexity-based cost-sensitive approach (termed ’iCost’) in this study. We first categorize all the minority-class instances based on their difficulty level and then the instances are penalized accordingly. This ensures a more equitable instance weighting and prevents excessive penalization.

Results: The performance of the proposed approach is tested on 65 binary and 10 multiclass imbalanced datasets against the traditional cost-sensitive learning frameworks. A significant improvement in performance has been observed (displayed in the figure below), demonstrating the effectiveness of the proposed strategy.

Paper - Arxiv

Code - GitHub

Machine Learning Enabled Multimode Fiber Specklegram Sensors: A Review

Multimode fiber (MMF) speckle-gram sensors have recently drawn significant attention due to the incorporation of ML algorithms in detecting different sensing parameters. Deep learning techniques provide an efficient way of extracting information from fiber specklegrams, which can be utilized in different sensing applications.

CNNs have proven to be extremely successful in imaging technologies over the last decade. The breakthrough from CNN has instigated new frontiers for applications in different domains. CNNs can automatically learn the variations in MMF specklegrams under different conditions. Besides detecting slight variations in the fiber, CNNs are also insusceptible to environmental noise and fluctuations, thus, making them superior in terms of performance accuracy. They provide a low-cost and simple alternative to extracting information from fiber specklegrams which has piqued the interest of many researchers.

In this project, the articles that explore the use of ML in different MMF sensing applications, such as bending sensors, endoscopes, tactile or position sensors, and others, have been reviewed. The principle of specklegram in MMF, the data generation process, an overview of related DL approaches as well as open challenges and future research directions in this field have been discussed.

The project is supported by the Deanship of Research Oversight and Coordination (DROC) at King Fahd University of Petroleum and Minerals (KFUPM) through the Center for Communication Systems and Sensing grant no. SB201008.

The paper has been published in the IEEE Sensors Journal.

Paper

An ML-based decision support system for reliable diagnosis of ovarian cancer by leveraging explainable AI

This work reflects the application of AI in medical diagnosis and the transparency that is required for a reliable decision-support tool.

Novel Biomarkers: Ovarian cancer (OC) is one of the most prevalent types of cancer in women. Early and accurate diagnosis is crucial for the survival of the patients. However, the majority of women are diagnosed in advanced stages due to the lack of effective biomarkers and accurate screening tools. While previous studies sought a common biomarker, our study suggests different biomarkers for the premenopausal and postmenopausal populations. This can provide a new perspective in the search for novel predictors for the effective diagnosis of OC.

Transparency: The stochastic nature of the ML algorithms as well as feature selection techniques raises concerns about the reliability of the system. To increase the trustworthiness and accountability of the diagnostic system as well as to provide transparency and explanations behind the predictions, explainable AI has been incorporated into the ML framework.

Results: The diagnostic accuracy obtained from the proposed system outperforms the existing methods as well as the state-of-the-art ROMA algorithm by a substantial margin.

Code

Paper