Research Interests:

1. Artificial Intelligence

2. Applied Machine learning

3. Computer Vision

4. Natural Language Processing

5. Data Science

6. Healthcare AI

7. e-Commerce Security

Current Research Projects:

Predicting Student Engagement in Collaborative Learning: In the Perspective of Bangladesh

A method of knowledge production based on the group is called collaborative learning. Psychosocial developmental theory suggests that student interactions in collaborative learning can enhance understanding and proficiency with key ideas. Traditional classes often neglect student feedback and engagement opportunities, whereas facial expressions and eye contact during collaborative learning can reflect students' performance. The study proposes predicting Bangladeshi students' collaborative learning participation through computer vision techniques. It suggests using a multi-modal deep neural network (MDNN) combining facial expression and gaze direction analysis to accurately predict performance and engagement levels. Predicting student engagement in collaborative learning environments using computer vision techniques among Bangladeshi students has significant socio-economic implications. It can improve education quality, optimize resource allocation, enhance employability, and guide evidence-based policymaking in Bangladesh.

Early Detection of Parkinson’s Disease Using Image Analysis

Parkinson's Disease (PD) was initially identified and documented by James Parkinson, an English physician, in 1817. It is characterized as a chronic, progressive neurodegenerative disorder that has a global impact, affecting individuals worldwide. The disease accounted for 5.8 million disability-adjusted life years in 2019, marking an 81% increase since 2000. These statistics emphasize the growing impact of PD on global health. In Bangladesh, every year, approximately 1600 people die from PD. Till now there is no cure for PD. However, about ten years before the onset of tremors or motor symptoms, the dopaminergic neurons begin to change. Some premotor symptoms of Parkinson’s disease (at an early stage) include a decrease in sense of smell, disorder in Rapid Eye Movement (REM) sleep, small handwriting, difficulty in moving, etc. This study aims to detect PD at an early stage utilizing image processing and convolutional neural networks. 

Improving Patient Care in Bangladesh through AI-Driven Handwritten Prescription Summarization and Medicine Category Prediction

Bangladesh, a South Asian low-middle-income economy, is witnessing demographic and epidemiological shifts due to rapid urbanization and increased life expectancy. As the seventh most populous country, it grapples with challenges related to non-communicable diseases (NCDs) and air pollution, leading to significant health concerns. However, healthcare accessibility and quality remain uneven, particularly affecting the underprivileged population. Handwritten prescriptions are a prevalent challenge, making prescription comprehension and medicine categorization difficult. To address this, we propose an innovative approach that leverages OCR-based deep learning for text extraction, followed by NLP for prescription classification and summarization, ultimately enhancing healthcare services in Bangladesh. This research carries significant relevance for Bangladesh's national development in the healthcare sector. By addressing the challenges posed by handwritten prescriptions and streamlining medicine categorization and summarization, the proposed AI-driven system aims to bridge healthcare disparities and elevate the overall quality of medical services in the country.

Completed Projects

Level Classification of Cognitive Load Using Unsupervised Machine Learning (Manuscript is in preparation for submission)

Cognitive load refers to the mental demand placed on working memory during a task. It makes use of an information processing model to explain how the mind gathers and stores knowledge as well as the constraints placed on working memory. Accurately classifying cognitive load levels has a great impact in various fields, including education, human-computer interaction, and workload management. This paper aims to classify the level of cognitive load using machine learning. The main objective of this study is to categorize the levels of cognitive load using characteristics taken from physiological signals such as electrocardiogram (ECG) and perceptual tasks. To do this at first K-Means algorithm, an unsupervised clustering model, is used and trained it by an augmented dataset for generating three clusters. By analyzing the perceptual and physiological features, these clusters are labeled as Low, Medium, and High levels of cognitive load. The results of this research will help to build dependable and effective machine learning techniques for classifying cognitive load.

AI-Enhanced Thyroid Disorder Diagnosis: Leveraging Routine Lab Tests in Bangladesh (Manuscript submitted to BMC Medical Informatics and Decision Making, currently under review)

Thyroid dysfunction is a leading endocrine disorder in recent times. Thyroid hormone excess and deficiency are frequently overlooked and misdiagnosed. In Bangladesh, around 10% of our population undergo clinically evident thyroid disorders. Recently, subclinical hypothyroidism and hyperthyroidism have been included as thyroid disorders adding another 10% population with thyroid dysfunction, totaling 20% of the population getting trouble with any type of thyroid disorder. The project aims to develop an explainable diagnosis support system using machine-learning algorithms to identify thyroid dysfunction with routine clinical data and demonstrate the potential to improve medical screening. 

Prediction of preeclampsia and associated risk factors in Bangladesh (Manuscript is in preparation for submission)

In Bangladesh, the incidence of preeclampsia is alarmingly high, with about 20% of maternal deaths associated with Preeclampsia/Eclampsia. It causes abortion, prematurity, intrauterine growth retardation, and stillbirth. Proper antenatal care remains an important part of prevention. Estimating the individualized risk of each woman allows antenatal surveillance to be directed at those women, who are most likely to develop preeclampsia. Such care leads to early diagnosis and intervention, both in terms of feto-maternal monitoring and timing of delivery. Hence, there is a need to develop an integrated model for the estimation of patient-specific risk factors for the development of preeclampsia based on maternal demographic, socioeconomic, obstetrics, nutritional, and anthropometric parameters. The use of machine-learning methods may assist in improving any prediction model and generate personalized risk assessment for the individual risk of a specific patient in the first trimester to develop preeclampsia. Therefore, this study aims to develop, train, and test an automated machine-learning model for the early prediction of suspected preeclampsia in pregnant women. 

Bangladeshi taka recognition for visually impaired persons (Manuscript is submitted to Applied Intelligence)

ChessAI: Playing chess like a human 

Credit card fraud detection using an ensemble-based soft voting approach in an imbalanced dataset (Published in Heliyon)