BRAC University Institutional Repository
Preserving Knowledge, Advancing Research, Sharing Scholarship
A digital platform for collecting, preserving, and sharing BRAC University’s scholarly, academic, and institutional outputs.
Contribute to the Repository
Faculty members and students are invited to submit their research publications, theses, dissertations, and scholarly works to increase visibility, access, and long-term preservation.
Recent Submissions
Efficient portfolio management using TOPSIS and ada-boost
(Institute of Electrical and Electronics Engineers Inc., 2020-12-16) Amanat Ullah, A.K.M.; Mahtab, Mohammad Tanvir; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
The nature of the stock market is random and uncertain and therefore it is difficult to make accurate decisions in stock trading. With this paper we propose a model which can select stocks effectively in the US stock market by feature extraction from data provided by the Quantopian platform. Our approach consisted of 17 features of 4 different domains. To determine the importance of each feature Ada-boost classifier was use. Then the topsis method was applied over 1500 stocks from the US stock market. After the applying the TOPSIS method Ideal solutions and Worst solutions were generated. Using those values all the stocks were given a performance score, which was used in selecting the stocks for the ideal portfolio. Our overall approach was to use Ada-boost to find the weights of each of the features and then apply TOPSIS to select the best stocks.
Bangla sign language recognition and sentence building using deep learning
(Institute of Electrical and Electronics Engineers Inc., 2020-12-16) Shurid, Safayet Anowar; min, Khandaker Habibul; Mirbahar, Md. Shahnawaz; Karmaker, Dolan; Mahtab, Mohammad Tanvir; Khan, Farhan Tanvir; Alam, Md. Golam Rabiul; Alam, Md. Ashraful; Department of Computer Science and Engineering
Modern age being the era of Information technology, it would not have come this far without the piled up data or information. Whereas communication is the basis of collecting or gathering data or information, almost 5% of the world's population is not blessed with the ability of verbal communication [1]. Sign language varies from the verbal language in every form and rule. This creates a gap between people conversing in verbal language and those communicating in sign language. Verbal languages are easy to interpret for having a common rule-following but sign language differs from region to region. This hampers the communication between normal people and those interacting in sign languages. Human to human interpretation is tough because of the enriched word wise signs and vocabs. To eradicate this issue, we are proposing a machine-based approach for training and detecting the Bangla Sign Language. Our aim is to create a multi modal system to for recognising Bangla signs. In addition, we hope to train the system with enough samples containing different signs used in Bangla Sign Language. In this research, we are using the Convolutional Neural Network (CNN) for training each individual sign. In addition to working as a medium of communication between the deaf and mute with the remaining society, this approach would also serve as a tool for the hearing deprived to learn and use the sign language properly.
Maternal nutrition intervention and maternal complications in 4 districts of Bangladesh: A nested cross-sectional study
(Public Library of Science, 2019-01-01) Todd, Catherine S.; Chowdhury, Zakaria; Mahmud, Zeba; Islam, Nazia; Shabnam, Sadia; Parvin, Musarrat; Bernholc, Alissa; Martinez, Andres; Aktar, Bachera; Afsana, Kaosar; Sanghvi, Tina; BRAC James P Grant School of Public Health
Background: Maternal morbidity is common in Bangladesh, where the maternal mortality rate has plateaued over the last 6 years. Maternal undernutrition and micronutrient deficiencies contribute to morbidity, but few interventions have measured maternal outcomes. We compared reported prevalence of antepartum, intrapartum, and postpartum complications among recently delivered women between maternal nutrition intervention and control areas in Bangladesh. Methods and findings: We conducted a cross-sectional assessment nested within a population-based cluster-randomized trial comparing a nutrition counseling and micronutrient supplement intervention integrated within a structured home-based maternal, newborn, and child health (MNCH) program to the MNCH program alone in 10 sub-districts each across 4 Bangladesh districts. Eligible consenting women, delivering within 42-60 days of enrollment and identified by community-level health workers, completed an interviewer-administered questionnaire detailing the index pregnancy and delivery and allowed review of their home-based care register. We compared pooled and specific reported antepartum, intrapartum, and postpartum complications between study groups using hierarchical logistic regression. There were 594 women in the intervention group and 506 in the control group; overall, mean age was 24 years, 31% were primiparas, and 39% reported facility-based delivery, with no significant difference by study group. There were no significant differences between the intervention and control groups in household-level characteristics, including reported mean monthly income (intervention, 6,552 taka, versus control, 6,017 taka; p = 0.48), having electricity (69.6% versus 71.4%, p = 0.84), and television ownership (41.1% versus 38.7%, p = 0.81). Women in the intervention group had higher recorded iron and folic acid and calcium supplement consumption and mean dietary diversity scores, but reported anemia rates were similar between the 2 groups (5.7%, intervention; 6.5%, control; p = 0.83). Reported antepartum (69.4%, intervention; 79.2%, control; p = 0.12) and intrapartum (41.4%, intervention; 48.5%, control; p = 0.18) complication rates were high and not significantly different between groups. Reported postpartum complications were significantly lower among women in the intervention group than the control group (33.5% versus 48.2%, p = 0.02), and this difference persisted in adjusted analysis (adjusted odds ratio [AOR] = 0.51, 95% CI 0.32-0.82; p < 0.001). For specific conditions, odds of retained placenta (AOR = 0.35, 95% CI 0.19-0.67; p = 0.001), postpartum bleeding (AOR = 0.37, 95% CI 0.15-0.92; p = 0.033), and postpartum fever/infection (AOR = 0.27, 95% CI 0.11-0.65; p = 0.001) were significantly lower in the intervention group in adjusted analysis. There were no significant differences in reported hospitalization for antepartum (49.8% versus 45.1%, p = 0.37), intrapartum (69.9% versus 59.8%, p = 0.18), or postpartum (36.1% versus 29.9%, p = 0.49) complications between the intervention and control groups. The main limitations of this study are outcome measures based on participant report, non-probabilistic selection of community-level workers' catchment areas for sampling, some missing data for variables derived from secondary sources (e.g., dietary diversity score), and possible recall bias for reported dietary intake and supplement use. Conclusions: Reported overall postpartum and specific intrapartum and postpartum complications were significantly lower for women in intervention areas than control areas, despite similar rates of facility-based delivery and hospitalization for reported complications, in this exploratory analysis. Maternal nutrition interventions providing intensive counseling and micronutrient supplements may reduce some pregnancy complications or impact women's ability to accurately recognize complications, but more rigorous evaluation is needed for these outcomes. Copyright © 2019 Todd et al.
Classification of non-topological magnetic configurations using machine learning
(Institute of Electrical and Electronics Engineers Inc., 2020-12-16) Bokul, Saffat; Shukur, Samiha Sabrin Md Abdus; Ahmed, Saquib; Bhowmick T.K.; Alam, Md. Ashraful; Department of Computer Science and Engineering
Skin cancer is a huge issue which gets neglected very often. Sometimes the human eye is unable to precisely detect diseases from imaging data, in cases of doctor's manual inspection. In this age, we see the rise of use of deep learning methods in our daily life problem solving. Therefore, we develop an automated computerised system for detecting skin diseases using deep neural network algorithms. In the proposed model, we have used several neural network algorithms and analyse their performances to detect five major skin diseases and Figure out the best performing algorithm in terms of accuracy. CNN and by using Keras Sequential API, we have structured a new model to gainan accuracy of around 80%. Later, for comparison and also to increase accuracy we have used architectures that use pre-trained data. These transfer learning model includes VGG11, RESNET50 and DENSENET121. Among the algorithms used in the proposed models, resnet architecture achieve highest accuracy of 90%.
LassoForest: CGPA dynamics through time-series forecasting and Counterfactual Analysis of Alcohol Consumption
(Institute of Electrical and Electronics Engineers Inc., 2026-01-01) Akter, Bushra; Tamim, Zannat Hossain; Rahat, Shahzalal Khan; Rahman, Md. Sazzadur; Department of Computer Science and Engineering
Predicting students' academic performance is one of the most important aspects of educational planning as it gives way to targeted interventions. This work suggest an temporal machine learning framework called LassoForest to forecast CGPA, which uses longitudinal behavioral data such as alcohol consumption. A sliding-window feature engineering method was used to track sequential trends in academic performance and lifestyle behaviors. The LassoForest model achieved an MAE of 0.80 and RMSE of 1.31, with a cross-validated R2 mean of 0.829 and standard deviation of 0.039. Counterfactual analysis simulating reduced alcohol consumption showed a slight increase in mean CGPA from 12.548 to 12.562, with minimal changes in standard deviation and percentile values. There were some students whose CGPA rose as high as 0.464 points. The findings imply that the combination of temporal behavioral patterns detection with an interpretable hybrid modeling approach allows one to not only make accurate CGPA predictions but also provide very useful insights for the case of behavioral interventions. The research seeks to prepare the ground for the educational stakeholders who wish to reduce the negative effects of alcohol on academic performance through targeted, data-informed behavioral counseling.