Browsing by Author "Islam, MD Saiful"

Now showing 1 - 3 of 3

Estimating flood susceptibility of Bangladesh in the future year using machine learning
(BRAC University, 2021-06) Alim, Sakib Bin; Lucky, Rakebun Islam; Ahmed, Aunindya Arif; Nahian, Prethu; Islam, MD Saiful; Syed, Shehran; Anik, Marum Monem
Being a riverine country with more than 400 rivers, flood is a common phenomenon for Bangladesh. As, the land is less than five meters above sea level, and also due to heavy rainfall during monsoon season, it makes the country an easy target of flooding and about 30% of the total area is in danger level during this period. Additional to the yearly flooding, every 4 to 5 years there is a major flood occurs which covers more than 60% of the country. As of 22 July, 2020 alone, 102 upazila and 654 unions have been inundated in flood, affecting 3.3 million people, leaving 731,958 people water logged and a total of 93 deaths [2]. The aim of this research is to predict Bangladesh’s susceptibility to flooding so that the government as well as the people of this country can take necessary steps to lessen the effect. To predict the probability of flood we will be using some machine learning algorithm namely Linear Regression model, Random forest Regressor, Naive Bayes Theorem and Artificial Neural Network. This study is based on the data set from 1991-2013 water level and weather variables from Khulna districts Rupsa-Pasur station.
Generative AI meets responsible AI and affective computing
(BRAC University, 2025-06) Eva, Atkea Fauzia; Shomrat, Kamran Hassan; Islam, Gazi Arman; Islam, MD Saiful; Subarna, Jamilatun; Alam, Md. Golam Rabiul
Generative AI, Responsible AI, and Affective Computing are transforming the future of artificial intelligence. The intersection of these fields represents a revolutionary breakthrough in computational technology. This thesis integrates these domains to develop a formalism for multidimensional emotional communication. By analysing image, voice, and text data, we address the challenge of detecting and generating emotions in real time, considering users’ gestures and interactions. We adopt an integrated approach based on deep neural network models across multiple modalities: text sentiment analysis, audio emotion detection, and facial expression recognition. In particular, we built our proposed approach using transformer-based models, including DistilRoBERTa, fine-tuned Wav2Vec2 on custom dataset, and DeepFace to process text, audio, and facial expression respectively. These pretrained models are trained for emotion classification with 6.7 million, 95 million, and 120 million trainable parameters, respectively. Natural Language Processing (NLP) models are used to interpret meanings and sentiments in text, while audio and image-based models detect emotional cues. The system adapts dynamically based on user feedback and incorporates Responsible AI practices such as bias detection, ethical safeguards, and safe interactions to ensure fairness and trustworthiness. Through practical experimentation and evaluation, we demonstrate that it is possible to build Generative AI systems capable of not only perceiving and reacting to human emotions but also generating emotionally appropriate responses. Potential applications include virtual assistants, mental health support tools, interactive storytelling systems, and educational platforms where enhanced emotional intelligence can significantly improve user experience.
Sentiment analysis using Natural Language Processing (NLP) & deep learning
(BRAC University, 2021-09) Islam, Kazi Minhazul; Reza, Md. Safkat; Yeaser, MD. Samin; Islam, MD Saiful; Rahman, Rafeed
It is an age of the Web and electronic media, and social media stages are one of the foremost frequently used communication mediums these days. But a few individuals utilize these platforms for a noxious reason and among those negative angles "Cyberbullying" is predominant. The way of monitoring user opinions throughout social media platforms such as Twitter and Facebook have been proven to be an e ective way of learning practically all of the consumers' thoughts which can open the door of potential future implementations. General emotion inspection can give us important data. The examination of supposition on informal communities, for example, Twitter or Facebook, has become an amazing method for nding out about the clients' sentiments and has a wide scope of utilizations. Notwithstanding, the productivity furthermore, the exactness of notion examination is being blocked by the di culties experienced in characteristic language handling (NLP). As of late, it is established that profound learning models are potential answers to the drawbacks of NLP. Natural language processing refers to a process that enables the machine to act like human and decreases the space between the person and the machine. Thus, NLP readily communicates with the computer in a straightforward sense. NLP has gained several uses in recent times. Each one of them are extremely e ective in daily life. An example can be a device which can be handled by voice commands. Several research workers are putting e ort on this idea in order to make even more real-life applications Natural Language Processing has tremendous potential to facilitate the use of computer interfaces for humans, as people will ideally communicate in their own language to the computer instead of learning an exclusive language based on computer instructions. In case of programming, traditional programming language's importance has always been underrated. This concept is questionable. We believe that modern Natural Language Processing techniques can make possible the use of natural language to express programming ideas, thus drastically increasing the accessibility of programming to non-expert users. Our team thinks that the implementation of natural language to convey programming concepts may be made possible by contemporary natural language processing techniques so that programming is accessible to inexperienced consumers substantially. The following paper surveys the most recent analysis that have utilized profound methodology how to take care of conclusion investigation issues, for example, assessment extremity. Models utilizing term recurrence opposite record recurrence (TF-IDF) and content insertion was implemented to an arrangement of datasets. At last, one similar examination of the exploratory outcomes was carried out in respect to several models and information highlights.