<?xml version="1.0" encoding="UTF-8"?>
<feed xmlns="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
<title>2026-2030</title>
<link href="http://suspace.su.edu.bd/handle/123456789/1634" rel="alternate"/>
<subtitle/>
<id>http://suspace.su.edu.bd/handle/123456789/1634</id>
<updated>2026-11-17T19:30:41Z</updated>
<dc:date>2026-11-17T19:30:41Z</dc:date>
<entry>
<title>Detection and Analysis of Cyberbullying  on Social Media Using Natural Language  Processing</title>
<link href="http://suspace.su.edu.bd/handle/123456789/3088" rel="alternate"/>
<author>
<name>Hossain, Alamin</name>
</author>
<id>http://suspace.su.edu.bd/handle/123456789/3088</id>
<updated>2026-11-10T05:46:05Z</updated>
<published>2026-09-10T00:00:00Z</published>
<summary type="text">Detection and Analysis of Cyberbullying  on Social Media Using Natural Language  Processing
Hossain, Alamin
The exponential growth of social media has amplified cyberbullying a pervasive form &#13;
of online harassment that inflicts severe psychological distress on victims and unlike &#13;
conventional bullying, cyberbullying on platforms such as Twitter operates through &#13;
rapid, public, and often anonymous interactions, rendering automated detection &#13;
particularly challenging. This study proposes a machine learning framework for &#13;
detecting and classifying cyberbullying in English social media text, utilizing a &#13;
balanced dataset of 10,000 manually annotated records across six categories (age, &#13;
ethnicity, gender, religion, other cyberbullying, and not cyberbullying) to train twelve &#13;
distinct models under standardized experimental conditions. Experimental results &#13;
demonstrate that a fine-tuned BERT model achieved the highest overall performance &#13;
with 84.65% accuracy, 84.59% precision, 84.65% recall, 84.48% F1-score, and 0.9734 &#13;
ROC-AUC, while Random Forest recorded the strongest baseline at 83.85% accuracy &#13;
and 83.94% F1-score; notably, BERT excelled at identifying identity-based harassment &#13;
with over 93% F1-score for age, ethnicity, and religion, whereas the linguistically &#13;
ambiguous "other cyberbullying" class remained the most challenging at 84.19% F1. &#13;
These findings confirm that transformer-based contextual representations significantly &#13;
outperform traditional classifiers for multi-class cyberbullying detection, and the &#13;
proposed framework offers a viable solution for assisting content moderators in &#13;
proactively identifying harmful content, thereby contributing to safer and more &#13;
inclusive online environments.
</summary>
<dc:date>2026-09-10T00:00:00Z</dc:date>
</entry>
<entry>
<title>A Comprehensive Study of Bengali Fake News  Detection Using Machine Learning, Deep Learning, and  Transformer-Based Ensemble Approaches</title>
<link href="http://suspace.su.edu.bd/handle/123456789/3087" rel="alternate"/>
<author>
<name>Afrin, Fariha</name>
</author>
<id>http://suspace.su.edu.bd/handle/123456789/3087</id>
<updated>2026-11-10T05:42:03Z</updated>
<published>2026-09-10T00:00:00Z</published>
<summary type="text">A Comprehensive Study of Bengali Fake News  Detection Using Machine Learning, Deep Learning, and  Transformer-Based Ensemble Approaches
Afrin, Fariha
The rapid spread of misinformation on social media and online news portals makes &#13;
automated fake news detection a critical research problem, especially for low-resource &#13;
languages like Bengali. This thesis presents a leakage-aware pipeline that compares ML, &#13;
DL, and Transformer-based Ensemble models on the same dataset to identify the most &#13;
accurate and reliable approach. Al though many fake news detection studies already exist, &#13;
most focus on English or use only one type of model. In contrast, this research focuses on &#13;
Bengali and compares Machine Learning, Deep Learning, and Transformer-based &#13;
Ensemble models on the same dataset. The main objective is to identify the most accurate &#13;
and reliable approach for detecting Bengali fake news by collecting and preprocessing &#13;
Bengali news, training different AI models, evaluating their performance, and selecting &#13;
the best-performing model. A total of 27 models, including classical ML, DL, &#13;
BanglaBERT, XLM-RoBERTa, and eight ensemble strategies, were trained using TF-IDF &#13;
and stylometric features. SHAP, LIME, and attention analysis confirmed that the models &#13;
relied on genuine content rather than artifacts. By avoiding data leakage, the results &#13;
provide a more trustworthy performance estimate, while dataset size and label noise, &#13;
rather than model architecture, remain the main limitations. Among all 27 evaluated &#13;
models, the Soft-Voting Ensemble achieved the best overall performance with 87.09% &#13;
accuracy, 88.79% F1-score, and 93.96% ROC-AUC. It outperformed all individual &#13;
models by combining the strengths of diverse Machine Learning, Deep Learning, and &#13;
Transformer-based models. In complex situations, the model predicts based on learned &#13;
patterns. Since no AI model is 100% accurate, it should be used as a decision-support &#13;
tool. Transformer models improve context understanding, and Ensemble methods &#13;
combine multiple models to improve accuracy, making Bengali fake news detection more &#13;
effective.
</summary>
<dc:date>2026-09-10T00:00:00Z</dc:date>
</entry>
<entry>
<title>A Lightweight YOLOv11n-Based Urban Surveillance  Framework for Real-Time Auto Rickshaw Detection  and Comparative Performance Evaluation Across  YOLO Versions in Dhaka City</title>
<link href="http://suspace.su.edu.bd/handle/123456789/3086" rel="alternate"/>
<author>
<name>Roy, Bipul Chandro</name>
</author>
<id>http://suspace.su.edu.bd/handle/123456789/3086</id>
<updated>2026-11-10T05:37:47Z</updated>
<published>2026-09-10T00:00:00Z</published>
<summary type="text">A Lightweight YOLOv11n-Based Urban Surveillance  Framework for Real-Time Auto Rickshaw Detection  and Comparative Performance Evaluation Across  YOLO Versions in Dhaka City
Roy, Bipul Chandro
Dhaka City experiences persistent traffic congestion due to increasing vehicle density, limited road &#13;
capacity, and the complexity of its mixed-traffic environment. Auto-rickshaws are a common &#13;
component of urban transportation, making their automated detection valuable for real-time &#13;
traffic monitoring and intelligent transportation systems. This study presents a lightweight &#13;
YOLO-based urban surveillance framework for real-time auto-rickshaw detection in Dhaka City. &#13;
An Auto-RickshawImageBD dataset containing 1,331 original images was used as the primary data &#13;
source. Through data augmentation using the Albumentations library, the dataset was expanded to &#13;
3,671 images. The images were annotated into two classes, namely auto_rickshaw and &#13;
non_auto_rickshaw, and prepared in the YOLO object-detection format. Three lightweight YOLO &#13;
models, YOLO11n, YOLO12n, and YOLO26n, were trained and comparatively evaluated using &#13;
Precision, Recall, mAP@0.5, mAP@0.5:0.95, and F1-score. The experimental results show that &#13;
YOLO11n achieved the best overall detection performance. On the test dataset, YOLO11n &#13;
obtained an mAP@0.5 of 83.93%, mAP@0.5:0.95 of 47.64%, Precision of 84.19%, Recall of &#13;
73.82%, and F1-score of 78.66%. YOLO26n achieved the second-best performance, while &#13;
YOLO12n produced comparatively lower detection results under the evaluated configuration. &#13;
Based on these findings, YOLO11n was selected as the primary model for the proposed lightweight &#13;
surveillance framework. The framework is intended to process traffic-camera imagery and provide &#13;
vehicle detection information that can support traffic-density estimation and congestion alerts. The &#13;
study demonstrates that a lightweight YOLO model can provide an effective balance between &#13;
detection accuracy and computational efficiency for auto-rickshaw monitoring in the &#13;
challenging urban traffic environment of Dhaka City.
</summary>
<dc:date>2026-09-10T00:00:00Z</dc:date>
</entry>
<entry>
<title>Advance Travel: A Modern Web-Based  Tourism Management and Automated  Booking Platform</title>
<link href="http://suspace.su.edu.bd/handle/123456789/3085" rel="alternate"/>
<author>
<name>Misty, Md. Amir Shorif</name>
</author>
<id>http://suspace.su.edu.bd/handle/123456789/3085</id>
<updated>2026-11-10T05:34:24Z</updated>
<published>2026-09-10T00:00:00Z</published>
<summary type="text">Advance Travel: A Modern Web-Based  Tourism Management and Automated  Booking Platform
Misty, Md. Amir Shorif
The tourism and hospitality sector have emerged as one of the most critical catalysts for &#13;
socioeconomic progress and cultural exchange globally and within Bangladesh. Despite &#13;
rapid digital adoption in recent years, travelers frequently encounter fragmented, &#13;
cumbersome, and manual reservation workflows characterized by opaque pricing &#13;
schemes, lack of real-time multi-modal scheduling, physical paper ticket reliance, and &#13;
cumbersome in-person payment procedures. Conversely, small-to-medium tourism &#13;
operators lack modern, affordable enterprise software suites capable of managing &#13;
dynamic tour inventory, automating electronic payment reconciliation, and authenticating &#13;
passenger tickets at boarding gates. &#13;
To overcome these pressing limitations, this undergraduate project presents "Advance &#13;
Travel," a modern, unified, web-based tourism management and automated booking &#13;
platform designed, engineered, and evaluated using modern software engineering &#13;
methodologies. Built upon the robust architectural foundations of the PHP 8.4 runtime &#13;
and the Laravel 12 Enterprise Model-View-Controller (MVC) framework, the system &#13;
bridges the operational divide between modern travelers and tour operators. The platform &#13;
provides a responsive, intuitive user interface developed using the Blade templating &#13;
engine and modern CSS design systems, allowing prospective tourists to effortlessly &#13;
explore domestic and regional destinations, discover customized tour packages (covering &#13;
bus, train, flight, and hotel options), configure customized travel dates and room &#13;
selections, and execute secure cashless transactions. &#13;
A pivotal innovation of Advance Travel is its automated multi-channel financial &#13;
settlement pipeline integrated with the SSL Comers national payment gateway. The &#13;
platform implements secure instant payment notification (IPN) webhooks, automated &#13;
hash-based checksum validation, and immediate transaction ledger recording, offering &#13;
instantaneous receipt generation and an optional cash-on-boarding mechanism. Following &#13;
transaction confirmation, the system dynamically synthesizes an encrypted, tamper&#13;
evident digital boarding pass embedded with a high-density Quick Response (QR) &#13;
verification token. A lightweight public verification sub-system allows station operators &#13;
and tour coordinators to instantly authenticate passenger boarding passes using &#13;
commodity camera-enabled devices without specialized hardware. &#13;
Furthermore, Advance Travel incorporates an enterprise administrative control dashboard &#13;
equipped with real-time operational analytics, interactive inventory CRUD controls, &#13;
booking lifecycle tracking, automated refund processing, and customer inquiry &#13;
management. Rigorous black-box, white-box, unit, and integration testing demonstrates &#13;
that Advance Travel satisfies stringent functional and non-functional engineering &#13;
requirements, delivering high availability, resilient transactional integrity, and total &#13;
immunity against prevalent OWASP Top 10 vulnerabilities including CSRF, SQL &#13;
injection, and XSS. The system demonstrates how modern software engineering standards &#13;
and green cloud computing principles can revitalize the digital tourism landscape.
</summary>
<dc:date>2026-09-10T00:00:00Z</dc:date>
</entry>
</feed>
