Browsing by Author "Rahman, Rafeed"

Now showing 1 - 20 of 89

A comparative analysis of deep learning and hybrid models to diagnose multi-class skin cancer
(BRAC University, 2023-05) Nawrin, Ishrat Nur; Trina, Tonusree Talukder; Rasel, Annajiat Alim; Rahman, Rafeed
Skin cancer is one of the most lethal and increasingly prevalent cancers in the world. Skin cancer develops when the epidermal (top layer of skin) cells divide abnormally, causing it to spread to other regions of the human body. Skin cancer exists in seven different varieties. The presence of malignant epidermal cells determines the type of skin cancer. Dermoscopy, spectroscopy, and imaging tests are primarily utilized to identify the malignancy. These procedures are expensive and prolonged. It may result in unfavorable effects such as bleeding, bruising, and infection as well. The narrow variances in multi class cancer pictures escalate the complexity of classification. Dermatologists confront challenges in the categorization of cancer types from images. Deep learning has resulted in a dramatic leap in disease identification. Deep learning models are capable of categorizing skin cancer more precisely than dermatologists. Several studies focused on pre-trained and hybrid models for categorizing the classes of skin cancer. In contrast to binary classification, the multi-class classification of skin cancer yielded an insignificant result for both deep learning and dermatologists. The proposed study employs varieties of deep learning and hybrid models to examine the performance of each model in categorizing the classes of cancer. The proposed CNN-SVM-LSTM hybrid model obtained the highest result compared to other models, with 87.15% accuracy, 87.42% precision, 87% recall, and 87.428% F1 score. To illustrate the overall comparison of the models, each model has been depicted through a classification report and a confusion matrix.
A comparative study of machine learning and geospatial techniques for analyzing Dengue diffusion patterns and identifying hotspots in Bangladesh
(BRAC University, 2025-01) Siddika, Taskia; Ethuna, Shuria Akter; Progga, Nafisa Ahmed; Ratul, Niamotullah; Kamal, Mirza Fahad Bin; Alam, Md. Golam Rabiul; Rahman, Rafeed
Dengue fever is still a major public health challenge in tropical and subtropical coun- tries, especially in Bangladesh where epidemic has been a big threat to public health. In this paper, we have developed an integrative computational approach analyzing geographic information and employing data mining to forecast dengue spread and reveal vulnerable regions. Our reference methods include Ordinary Least Squares (OLS), Geographically Weighted Regression (GWR), Lasso Regression, Elastic Net Regression, Bidirectional Long Short Term Memory (BiLSTM), and DiffFlow as well as TabDDPM. The study builds on past results, climatic parameters, and population density to improve predictive performance. The data set used in this research was collected from the official web site of Directorate General of Health Services (DGHS) that made the data authentic. The proposed approach, as a result, provides sig- nificantly higher predictive performance than conventional statistical analysis based on the spatial and machine learning components. Furthermore, to categorize and prioritize the high-risk areas, we apply special methods of multiple criteria decision making – TOPSIS and VIKOR. We reveal that the proposed models based on ma- chine learning methodologies are useful for identifying dengue fever hotspot areas and enlightening information for public health officials regarding timely application of control measures.. This study emphasises the necessity of the epidemiological and climate data coupled with the computational modeling to mitigate future outbreaks."
A comprehensive hybrid framework for Parkinson’s disease detection: integrating handcraft features along with deep learning-based feature extraction with variational autoencoder and traditional machine learning techniques for classification
(BRAC University, 2024-10) Alam, Md. Iftiajul; Laiba, Faria Islam; Nazi, Tahiatun; Choudhury, Shirsadip; Alam, Md.Golam Rabiul; Rahman, Rafeed
Neurodegenerative disorders, such as Parkinson’s disease, present a significant medical challenge, necessitating innovative approaches for detection. This thesis introduces a comprehensive hybrid framework that combines handcrafted features and deep learning techniques to improve the accuracy of Parkinson’s disease detection. The approach leverages pre-trained convolutional neural networks (CNNs) such as VGG16, MobileNet, and EfficientNet, ResNet to extract features of melspectrograms generated from the voice samples. A second contribution is the extraction of handcrafted features from the raw audio data. The features extracted are encoded using a Variational Autoencoder (VAE), which further reduces the dimension and integrated them to further train the machine learning algorithms such as Random Forest Classifier (RFC), K-Nearest Neighbour (KNN), Logistic Regression (LR), Support Vector Machine (SVM), and XGBoost to differentiate. To achieve this, we leveraged the combined strengths of these models by integrating both handcrafted and deep learning features to construct a highly optimized and effective classification model using a hybrid approach that highlights the potential of feature extraction techniques and advanced machine learning algorithms for improving the detection and diagnosis of Parkinson’s disease and facilitating more progress in computational healthcare and early stage diagnostics.
A deep learning approach for automated classification of Corneal Ulcers
(BRAC University, 9/25/2023) Barua, Sumit; Saha, Samit; Bulbuli, Jannatul; Rahman, Akib; Fuad, Md.Ibna Salam; Rahman, Rafeed; Nahim, Nabuat Zaman; Dofadar, Dibyo Fabian
Eye Corneal Ulcer(ECU) has been demonstrated to be the second most common cause of treatable blindness worldwide, after cataracts. It is an extremely prevalent ophthalmic ailment and can cause severe visual impairment or perhaps total blindness. This thesis renders a comprehensive study on the automated classification of corneal ulcers using a deep learning approach. In this research, the SUSTech-SYSU dataset has been utilized which is obtained from Sun Yat-sen University’s Zhongshan Ophthalmic Center, consisting of 712 images of patients with various types, grades and categories of corneal ulcers. These ocular surface images are captured after fluorescein staining, aiding as a valuable resource for the enhancement of deep learning models. The images in the dataset having dimensions of 2592 pixels in width and 1728 pixels in height, depicts close-up views of corneal abrasions under cobalt blue light during eye examinations, which is the particular type of image captured in this dataset. This thesis occupies the deep learning Convolutional Neural Networks (CNN) architecture, which includes InceptionV3, ResNet50 and VGG16 in order to create a pre-trained model for Eye-Corneal-Ulcer (ECU) image classification. In addition, a customized model is built to foster validation and test accuracy. The deep learning models are run on training and testing sets, enabling them to recognize unique criteria and patterns linked with different types of corneal ulcers. For data training, the dataset has been allocated by dividing it into separate folders based on the type of ECU images. Data augmentation is exerted by using the ImageData- Generator to escalate the diversity of the dataset and improve model generalization. The dataset comprises 10,000 training images and 2,000 testing images for evaluating the model. In the customized model, a sequential architecture is implemented, including layers such as Conv2D, max pooling, batch normalization, flatten, and dense layers for feature extraction and classification. For multi-class classification, categorical cross-entropy is employed as the loss function, and the Adam optimizer is used. Hyperparameter tuning has been enacted using the validation set, encompassing various learning rates, batch sizes, and regularization techniques to optimize the performance of the model. The consequence of this research avails the development of automated corneal ulcer classification, potentially facilitating ophthalmologists in diagnosing and curing corneal infections more productively.
A deep learning approach for multi-class bus fitness classification using a modified faster R-CNN model
(BRAC University, 2025-10) Khurshid, Fahim; Islam, Samiha; Rahman, Mohammed Raqin; Nusrat, Sadia; Hassan, Arif; Dofadar, Dibyo Fabian; Rahman, Rafeed
The high rates of development of the public transportation systems have caused the necessity of the creation of a stable, scalable, and automated system to check the vehicles in order to eliminate additional risk to the passengers and reduce the costs of their maintenance. This paper will present a deep learning driven architecture that uses a customized Faster Region Based Convolutional Neural Network (Faster R-CNN) to classify the bus fitness into multiple classes and hence removes the timeconsuming, inaccurate, and subjective task of manually examining structural defects, body states, and missing parts including side mirrors and headlights. In contrast to the traditional Faster R-CNN models which employ the use of the standard region proposal networks (RPN) and fully connected heads, our design features specialized Multi Layer Perceptron (MLP) heads to enhance feature segregation in subtle defect classes. Pre-processing and augmentation strategies also enhance the methodology by providing resistance to noise and change of viewpoint. This paper further extends the Faster R-CNN architecture of vehicle inspection by tackling the domain-specific limitations, such as small objects of defect and high intra class similarity, and class imbalance, by applying MLP-based classification heads and transfer learning with pre-trained weights, and complementing this with anchor refinement to enhance localization performance. Through experimental tests which include the mean average precision and the recall curves and the confusion matrices, significant gains are achieved compared to the baseline models especially with small or partially visible defects. These works include an expansion of object detection algorithms to safety critical applications, demonstration of the usefulness of feature space expansion using multilayer perceptrons, and the future prospects of implementing these algorithms in roadside camera devices and depot inspection systems, thus providing a base to smart transportation systems that can be applied to trucks, trains and aircraft.
A deep learning approach to predict crypto-currency price by evaluating sentiment and stock market correlations
(BRAC University, 2023-01) Maliha, Miftahul Zannat; Trisha, Ananya Subhra; Tamzid Khan, Abu Mauze; Das, Prasoon; Shakil, Shuhanur Rahman; Hossain, Muhammad Iqbal; Rahman, Rafeed
For the technological shift, advancing epoch towards cryptocurrency intensified the impactful method. Metaverse can originate the base operation into a diversified level. The extension of digital marketing contributes to blockchain technology more.Our research demonstrates, attested cryptocurrency price evaluation associated with the stock and sentiment. In our research, we have implemented various techniques to predict cryptocurrency prices. Crypto like bitcoin, ethereum and litecoin are the primary focus in this paper. Our research observes the fluctuation into the cryptocurrency prices. In our research procedure, we used the LSTM-GRU hybrid, ARIMA for time series prediction. The research follows sentiment analysis from the twitter scrapped data. The research provides cogent insights of cryptocurrency price prediction fluidity with the stock price and the twitter sentiment on following cryptocurrencies. Additionally, the data merge with the LSTM time series model depicts the cryptocurrency stock market and shows us the relationship between stock price, twitter sentiment and cryptocurrency price pertinence
A portfolio of BeaconTech Limited
(BRAC University, 2025-04) Nijhum, Nazia Ahmed; Rahman, Rafeed
Web Development is the process of designing websites and web applications. It requires coding,maintaining websites and ensuring that they are user-friendly and easy to use on different devices. The frontend development layer is accountable for validating the website’s functionality. It also ensures the website’s functionality as it helps to align with user-centric design principles as well as interactive requirements. Frontend usually employs core technologies such as HTML for structural integrity, CSS for visual styling and responsive layouts and JavaScript to enable dynamic behavior and seamless user engagement. Backend development layer utilizes server side programming language related stuff. These include PHP, Python, Ruby and JavaScript. It also focuses on managing underlying business logic, data processing and seamless integration with the frontend as well as the DBMS structure. It helps to ensure robust functionality , Scalability and secure data handling for the website. Basically, this paper elaborates my internship experience at Beacontech Limited.It is a sister concern of Optimum Solution and Services Ltd.In my internship period, I have participated in projects relating to my field and also I have learned insights into project management, ERP systems and corporate advancement tactics.The report also features the company profile, SWOT analysis and an in-depth perspective of the Beacon ERP system that enhances operations for businesses worldwide.
A predictive analysis of chronic kidney disease using machine learning
(BRAC University, 2022-09) Khan, Md.Shafayet; Afrida, Nazihan; Rahman, Munia; Islam, Sujana; Banik, Ananya; Rahman, Rafeed
Chronic kidney disease (CKD) is a determined disease condition having critical grimness and death rate that influences the whole grown-up populace brought about by either renal pathology or diminished renal capabilities. Early location and powerful treatments might have the option to end or diminish the growth of this constant condition to last stage, where dialysis or kidney transplantation is the main life-saving choice for patients. In this examination, we have investigated the opportunities for early chronic kidney disease expectation utilizing an assortment of machine learning algorithms. Here, a reasonable CKD dataset was taken from Tawam Clinic in AlAin city (Abu Dhabi, Joined Middle Easterner Emirates). We have proposed Support vector machine (SVM), Random forest algorithms (RF), Logistic regression (LR), Multinomial naive bayes (MNB), LSTM and contrasted their results with figure out the best exactness among the models. As a result, the models yielded outstandingly great order precision, with a LSTM exactness of 0.95 percent. The result of the review shows that improvements in machine learning (ML), with the assistance of prescient knowledge, comprise a reasonable climate for recognizing commonsense arrangements, which thus exhibit the prescient capacity in the space of renal illness and then some.
A secured federated learning system leveraging confidence score to identify retinal disease
(BRAC University, 2023-05) Eshan, M Sakib Osman; Nafi, Md. Naimul Huda; Sakib, Nazmus; Maruf, Md. Ahnaf Morshed; Emon, Mehedi Hasan; Reza, Tanzim; Rahman, Rafeed; Parvez, Mohammad Zavid
Federated learning is a distributed machine learning paradigm that enables multiple clients to collaboratively train a global model without sharing their local data. How- ever, federated learning is vulnerable to adversarial attacks, where malicious clients can manipulate their local updates to degrade the performance or compromise the privacy of the global model. To mitigate this problem, this paper proposes a novel method that reduces the influence of malicious clients based on their confidence. We conducted our experiments on the Retinal OCT dataset. The proposed technique significantly improves the global model’s precision, recall, F1 score, and area under the receiver operating characteristic curve (AUC-ROC). Precision rises from 0.869 to 0.906, recall rises from 0.836 to 0.889, F1 score rises from 0.852 to 0.898, and AUC-ROC rises from 0.836 to 0.889.
An AI and NLP approach for detecting grooming behavior
(BRAC University, 2024) Shanto, Hasibul Hossain; Farooqui, Farhan; Rafi, Abdullah Al; Feona, Maisha Maliha; Phul, Progya Talukder; Alam, Md. Golam Rabiul; Rahman, Rafeed
"Grooming children on social media is a dangerous side effect of modern internet era. AI models, specially NLP have the potential to play a critical role in detecting grooming behavior. Even though, there have been studies in the past to build a grooming detection system, there is limited research on building such systems us- ing modern NLP techniques. In this paper, we propose a modern sexual grooming detection system using state-of-the-art NLP models and techniques that can detect and alert users to potentially dangerous online interactions between groomers and their targets. Our detection system is a ConversationClassifier which is able to clas- sify conversations, whether they are grooming or not. With over 19,000 grooming sentences collected from PervertedJustice grooming conversations, we created an annotated dataset exhibiting the grooming characteristics. Conversational data was also collected from both PervertedJustice and PAN12 dataset. With the sentence- level annotated dataset, we trained a SentenceClassifier model based on RoBERTa & DeBERTa to be able to accurately predict if a sentence has grooming character- istics or not. The ConversationClassifier was built on top of the SentenceClassifier with LSTM & GRU to capture the sequential features in the conversation. Further- more, a self-attention mechanism was added so that the model can focus on relevant sentences. Our models achieved promising results. In case of the SentenceClassifier, it displayed an accuracy of 93% for RoBERTa and 94% for DeBERTa. We paired the RoBERTa based SentenceClassifier with LSTM which yielded an accuracy of 97% and DeBERTa based SentenceClassifier with GRU which yielded an accuracy of 95%."
An approach to detect epileptic seizure using XAI and machine learning
(BRAC University, 2022-05) Bijoy, Emam Hasan; Rahman, Md. Hasibur; Ahmed, Sabbir; Laskor, Md. Shifat; Hossain, Muhammad Iqbal; Rahman, Rafeed
One of the most common neurological disorder in health sector is Epileptic Seizure (ES) which is occurred by sudden repeated seizures. Hitherto more than 50 million people in the whole world are suffering from Epileptic Seizures. The abnormal brain activity of the central nervous system often causes unusual behavior, losing awareness and psychological problems etc. Moreover, many risks associated with epileptic seizures include sudden unexpected death in epilepsy (SUDEP) which is really a concerning problem discussed in this article. For abstaining from adverse consequences of epileptic seizure-like this health sector focuses more on the early prediction and detection of epilepsy. The complex signals of brain activity are reflected as swift-passing exalted peaks in Electroencephalogram (EEG). Initially, the specialist inspects the EEG signals over a few weeks or months to identify the presence of epileptic seizures, which is a very time-consuming and challenging task. Hence, Machine learning (ML) based classifiers are capable to categorize EEG signals and detect seizures along with displaying related perceptible patterns by maintaining accuracy and efficiency. In order to detect epileptic seizures, EEGbased signal recognition algorithms had been shown in this paper by applying both Multi-Class Classification and Binary classification. The algorithms were Decision Tree Algorithm, Random Forest Algorithm, Multi-Layer Perceptron (MLP) and K-Nearest Neighbor (KNN), Gradient Boosting Classifier, Gaussian Na¨ıve Bayes, Complement Na¨ıve Bayes, SGD Classifier, Explainable Artificial Intelligence (XAI), LIME Algorithm etc. However, K-Nearest Neighbor appears with pretty higher accuracy in certain conditions.
An approach to detect smartphone addiction through activity recognition and app usage behaviour
(BRAC University, 2023-05) Uz Zaman, Nur; Akther, Afroza; Tabassum, Nowshin; Samrat, Md. Khaliduzzaman Khan; Khan, Swad Mustasin; Alam, Md. Golam Rabiul; Rahman, Rafeed
The widespread use of smartphones has raised concerns about problematic smartphone use or addiction, which has become a significant issue in today’s society. Despite the recognition of this research area, detecting smartphone addiction remains a challenge. Therefore, it is crucial to identify the primary causes of smartphone addiction and understand how individuals’ lifestyles contribute to this behavior. Most of the methods in research area are self assessment based and detected via different addiction scales. Moreover, in previous studies daily human activities was never considered as a factor in problematic smartphone use. This study aims to explore a new approach in detecting excessive smartphone usage by considering the impact of sensor based daily activities and smartphone app usage. By examining addictive characteristics of smartphone usage and clustering them based on various independent variables, we sought to determine smartphone addiction and investigate the influence of daily activities. To collect reliable and accurate data, we utilized apps for seven days to capture information on the participants’ smartphone usage. Leveraging sensor data and LSTM models, we identified participants’ activities and correlated them with daily app usage duration to detect smartphone addiction using clustering methods such as K-Means and K-Medoids. Our analysis revealed that around 28% participants showed addicted behaviour. To validate these findings, we compared our result with survey results using diverse evaluation metrics (RI,FMI), which exhibited 87% accuracy.
An efficient deep learning approach to detect neurodegenerative diseases using retinal images
(BRAC University, 2023-01) Irfanuddin, Chowdhury Mohammad; Shafin, Wasique Islam; Ahmed, Koushik; Khan, Md. Hasib; Md. Ashraful, Alam; Rahman, Rafeed
Neurodegenerative disorders are diagnosed through undergoing brain MRI, CT scans, genetic testing, and various laboratory screening tests which are often tedious, time consuming and beyond the means of most people’s financial capabilities and sometimes health unconducive. To remedy this, we proposed an efficient deep learning approach to detect neurodegenerative diseases, for instance, Multiple Sclerosis, Parkinson’s disease, Amyotrophic Lateral Sclerosis, and Alzheimer’s disease using retinal images. Efficient convolutional neural network-based architectures are used to classify brain diseases. The system enables the detection of brain diseases from retinal images rather than brain images effectively. Through the proposed system, we are able to proactively detect such disorders simply through retinal scans which are faster and simpler compared to the scanning of the brain itself which requires expensive and sophisticated equipment. We conducted our research on a dataset containing retinal cross-sectional images of 21 Multiple Sclerosis patients and 14 healthy individuals. Our model achieved 100% accuracy in classifying all healthy and diseased individuals from retinal scans.
An end-to-end framework for anomaly detection and categorization
(BRAC University, 2025) Islam, MD. Farhan; Islam, Rehnuma; Reza, Syed Rahin; Tasnim, Saifa; Nipu, Anipa Akter; Rahman, Rafeed
In this study, we proposed an end-to-end framework for anomaly detection, classification in Industry 4.0 using deep learning models YOLO V8 and ResNet on the MVTec Anomaly Detection(MVTec AD) dataset. The framework is based on defect detection, anomaly localization. The multitask queues in YOLO V8 guarantee both: fast and precise detection in real time, while ResNet primarily suited for classification, complete with top notch precision and recall metrics. The metrics used for evaluation (including AUC, accuracy, precision, recall, F1 score and AP) confirm the good performance of the models. We also provide decision surface visualizations through Grad-CAM and Integrated Gradients that will help you understand some of the decisions made by the model. The YOLO V8 performed optimal on real-time detection tasks and ResNet performed best on classification accuracy, as highlighted through the results. This framework allows for the automation of anomaly detection and the resolution through investigation, unlocking future opportunities for real time anomaly detection and management.
An internship report on animation and software development
(BRAC University, 2025-02) Mostafiz, Shaolin; Rahman, Rafeed
I completed my internship at TechnoMagic Private Limited, an emerging IT solutions organization in Dhaka, Bangladesh. The company mostly specializes in software development, animation, and game development. Acquired hands-on experience on real projects along with animation principles and project management. This internship at TechnoMagic Company gave me an excellent grasp of animationlevel operations, enabling me to experiment with fundamental approaches in motion design, character movement, and visual narrative. This work experience helped me realize my newly found interest in the technical parts of animation as well as digital marketing, which drove me to learn more about the area. While I initially concentrated on learning the fundamentals, my internship acted as a springboard, providing me with a solid foundation in animation concepts as well as an awareness of the creative and technical collaboration necessary in the business. This internship experience not only taught me the very foundations of animation but also expanded my knowledge of how animation works with technology-driven solutions.
Analyzing Schizophrenic-prone text from social media content: a novel approach through ML and NLP
(BRAC University, 2024-01) Rodela, Raisa Rahman; Efty, Farhan Tanvir; Rahman, Mubashira; Wajiha, Shaira; Reza, Md Tanzim; Rahman, Rafeed
Schizophrenia is one of the destructive personality disorders where people have unusual interpretations of reality and are lured to develop harmful actions if not diagnosed promptly. This study focuses on identifying language patterns indicative of schizophrenic-prone texts in online communication and intends to contribute to the development of early intervention techniques in mental health utilizing ML and NLP methods. This study used two datasets to examine language patterns associated with schizophrenia in social media posts. The first dataset, Pre existing obtained from a repository focused on identifying schizophrenia-related postings, functions as a standard for comparison and evaluation. The second dataset, New scrapped obtained by extracting information from subreddits associated with schizophrenia, offers a more extensive range of language patterns. The dual-phase technique entails training models using the existing dataset and evaluating their performance on the newly collected dataset. The research uses various models, including transformer model BERT, recurrent neural network model Bi-LSTM, and GRU, as well as machine learning models such as Support Vector Classifier, Logistic Regression, Multinomial Naive Bayes, Random Forest, and Decision Tree to predict whether textual data is suggestive of schizophrenia. The language patterns of schizophrenic-prone texts differ from texts written by mentally-healthy individuals, encompassing phonological, morphological, and syntactic aspects. These models can analyze linguistic patterns and acquire knowledge about them. The results achieved after the training of the models are outstanding. The DistilBERT transformer model achieves 97% and 84% accuracy, GRU achieves high accuracy rates of 91% and 79%, the logistic regression machine learning model demonstrates impressive efficiency with accuracy rates of 93% and 83% respectively for Pre existing and New scrapped dataset. In order to ensure the models can effectively handle new data, we conducted a contemporary comparison. This analysis revealed that consistent data collection is necessary for accurate predictive results.
Analyzing the security differential privacy provides and the trade-off between performance and privacy in medical image classification
(BRAC University, 2024-05) Haque, Sumaiya; Mehraj, Mohammad Azim; Rahman, Mohammad Faiazur; Abedin, Mahmud; Reza, Md Tanzim; Rahman, Rafeed
One of machine learning’s main purposes is to draw out functional and practical information from a set of data while perpetuating the entire privacy by protecting all information. While it might seem a bit hard to maintain, privacy does play a vital role in every sector, and thus, the information must be frequently balanced, especially when extracting sensitive datasets. For instance, medical research or image classification can be considered an important application where patient privacy, as well as the extraction of information, are both of utmost importance [12]. Medical images are details that consist of a patient’s private information and are collected from various hospitals, nursing homes, and research institutes. Later on, these images are utilized to infer a patient’s physical condition, ultimately leading to an invasion of privacy[10]. In recent years, medical images have become a prominent research and analysis subject, and therefore more and more people are getting affected as their private information is being shared. Thus, in our research, we are going to showcase different ways to defend against information leakage. Differential privacy is considered one of the strongest forms of privacy because we work with privacy-preserving algorithms and learning-based mechanisms. Apart from that, federated learning and image watermarking can also help in preserving privacy. Deep learning techniques that can be utilized to preserve data utilizing Conditional GANs also face particular difficulties when used with medical images. In order to show the optimal method of data preservation, we will attempt to collect a dataset.
Attention-deficit/hyperactivity disorder detection leveraging an ensemble of encoder-decoder transformer and XGBoost models
(BRAC University, 2024-10) Sarker, Sharon Rose; Mehjabin, Saowmi; Piper, Meherin Majid; Rahman, Rafeed
"Early detection of neurodevelopmental disorders such as Attention-Deficit/ Hyperactivity Disorder(ADHD), can lead to improved outcomes and prompt intervention. Traditional detection methods have been facing challenges due to judgment and misinterpretations, lack of resources, and biasness which may cause under-diagnosing or over-diagnosing. Early detection of these neurodevelopmental disorders, not only helps individuals to get proper ministrations but also it can improve their social, cognitive and mental development. In this study, our aim is to build an ensemble model leveraging a custom Transformer with various attention mechanisms alongside an XGBoost model to improve diagnostic accuracy. By comparing the proposed model with other traditional machine learning and deep learning models, this study aims to enhance the accuracy and efficiency of diagnosis. By using a pre-processed EEG dataset and customized ensemble model, the proposed model has achieved 83% of accuracy, highest accuracy among the traditional models. Moreover, this research aims for future development in the field, by offering methodologies that can be useful to further studies focused on disorder detection. In conclusion, this research will use an ensemble model leveraging a custom Transformer with various attention mechanisms alongside an XGBoost model for early ADHD aiming to create a new precision and accessibility in identifying neurodevelopmental disorders. "
Automated reference validation for scholarly publications using NLP
(BRAC University, 2024-01) Khan, A S M Nasim; Khan, Mohammad Nasif Sadique; Howlader, MD. Adnan; Roy, Ayan; Alam, Md. Golam Rabiul; Sadeque, Farig Yousuf; Rahman, Rafeed
Accurate references in scholarly publications are a crucial aspect of scientific writing. The manual validation of references can be a time-consuming and error-prone process. This research introduces an updated version of the automated referencing validation model that makes the peer review process efficient. The proposed model utilizes the capabilities of Natural Language Processing generating sentence embeddings which uses an efficient algorithm. Our model first breaks down the scholarly article into sections and uses topic modeling to group every section according to their context properly. After that, It generates sentence embeddings for each section. By making sets of embeddings, they are used to calculate the semantic similarity between the query and the referred article. Additionally, this methodology addresses the valid references for non-contextual scenarios such as having common name entities. Lastly, strategic feature engineering is also being used for better performance. We have created a dataset of scholarly papers with manually verified references to evaluate the efficiency and accuracy of our model. This improved version of the referencing validation model aims to outperform traditional models such as Document-BERT, BERT, and SBERT regarding efficiency and accuracy. The model can be used in interactive real-time systems, providing quick and reliable feedback to peer reviewers. This study aims to make a contribution to the field of automated referencing validation in scholarly publications. The model offers a solution to the limitations of manual validation which makes it a valuable tool for peer reviewers and researchers.
Brain hemorrhage detection using hybrid machine learning algorithm
(BRAC University, 2022-01) Iqbal, Khondoker Nazia; Azad, Istinub; Emon, Md. Imdadul Haque; Amlan, Nibraj Safwan; Aporna, Amena Akter; Islam, Md. Saiful; Rahman, Rafeed
Machine learning (ML) helps computers learn and program data without humans’ help. According to data scientists, machine learning can extract 60% high-quality information, reduce the cost up to 46%, and increase operation speed by approximately 48% [1]. Recently, there has been successful implementation of machine learning in data analysis, computer vision, computer-aided diseases (CAD), and many more fields. Machine learning is broadly used in the medical industry because of its processing power for image data and pattern recognition quality. The image processing power of machine learning can be used in medical images to classify the brain images automatically. Segmentation and classification of brain image can provide valuable information and quantitative assessment of lesions which can be used for treatment strategies and predicting patient condition (Kamnitsas et al., 2017). According to research [2], an estimated 64-74 million people in the world are affected by traumatic brain injury every year. It affects the lives of nearly every one out of six persons. In our proposed system, we will use a hybrid approach of multiple machine learning algorithms together for the classification of CT brain images and diagnose brain disorders and diseases like brain hemorrhage. Some ML algorithms such as different 3D Convolutional Neural Networks (CNN) , AlexNet, DenseNet121, GoogleNet and some other models like Multilayer Perceptron Model (MLP), Support Vector Machine (SVM) and Random Forest (RF) have been applied successfully in this field in the past. Modifying previous methods, we want to build a hybrid machine learning algorithm by combining different CNN models like VGG-16, VGG-19, Random forest and Multilayer Perceptron (MLP) classifiers for detecting brain hemorrhage. We have used the VGG-16 and VGG-19 model to derive image features from the CT brain images and Random forest classifier and MLP classifier for testing the accuracy of our model. To test the efficiency of our system, we have used CT brain image datasets from Kaggle. The CT brain imaging data will be the input of our model and our model will detect brain hemorrhage and classify them into one of six classes: Epidural, Intraparenchymal, Intraventricular, Subarachnoid, Subdural and No Hemorrhage. Using our hybrid approach the best accuracy we achieved was around 97.24% using a combined approach of VGG-16 and Multilayer Perceptron classifier. Also we used Explainable AI to explain the prediction of the hemorrhagic classes.