http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17240 2026-07-21T05:11:07Z http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17467 Prediction Of Diabetes Using Machine Learning Algorithms Rafsun, Sk. Salman Diabetes is a frequent condition in humans that is brought on by a collection of metabolic diseases in which the body's sugar levels remain abnormally high for an extended length of time. Because it damages many of the body's systems by affecting various organs, we are all trying to prevent diabetes at an early stage by anticipating its symptoms using a variety of techniques. Human life can be saved by controlling such diseases early on. In order to accomplish the goal, this research project primarily uses machine learning approaches to investigate differentrisk variables associated with this disease. Effective knowledge extraction is accomplished by machine learning methods that build prediction models using diagnostic medical datasets from diabetic patients. It may be possible to forecast diabetic people by gleaning information from such data. K-Means Cluster, Naive Bayes (NB), Support Vector Machine (SVM), K- Nearest Neighbor (KNN), Linear Regression, Decision Tree (DT), Logistic Regression, Random Forest (RF) and Hierarchical Cluster are nine well-known machine learning algorithms that I use in this work to predict diabetic disease using data from the adult population. In comparison to other machine learning methods, the findings from my experiments indicate that when compared to alternative approaches, the C4.5 decision tree attained a greater level of accuracy. M.SC. in ETE 2025-01-25T00:00:00Z http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17448 A Hybrid AI-Based Model for Depression Detection in Bangla Social Media Posts Using the BSMDD Dataset Roy, Pijush Chandro Depression is a significant healthcare problem worldwide, impacting countless people of all ages annually. This paper presents a novel approach for depression identification by using complex datasets and employing various machine learning models such as LightGBM, XGBoost, Naïve Bayes, and Random Forest. The study emphasizes how itis possible to change the face of mental health diagnosis with those models relying on the data. In the future, the analysis should aim at improving model interpretability, reducing bias in the algorithms, using multi-source data as well ethical issues such as rights of privacy and consent. These innovations are designed to help improve the diagnostic capabilities, promote inclusiveness in the outcomes, and support the means of real-time monitoring their health of the individual. Among the models tested, LightGBM proved to be the best as it achieved an accuracy of 83.67% and an F1 score of 84.06%. XGBoost had a higher accuracy rate while Naive Bayes had a higher recall rate. Random Forest also proved effective, performing well in all areas metrics which show their different benefits for dealing with issues in mental health care. M.SC. in ETE 2025-10-15T00:00:00Z http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17438 Machine Learning Based Resource Allocation Optimization in 5G Network Shourov, Mohammad Shakib Hosen The Fifth-generation (5G) wireless represents a significant advancement for mo- bile connectivity successfully, near-instant response times, the ability to connect a huge number of devices reliably. But this power comes with a catch: managing its radio resources is incredibly complex, thanks to constantly shifting network con- ditions and wildly different user demands. How can the vital resources of a 5G network effectively managed? Question by applying machine learning techniques to the problem explored by this thesis. In order to replicate both the core and radio access network, started by creating a simulated 5G environment with open-source platforms Open5GS, Free5GC, and UERANSIM. with the testable place, we are able to systematically trial and analyze various resource allocation methods. To improve resource allocation performance, were applied some machine learning ap- proaches as like Linear Regression, Polynomial Regression, and XGBoost. In order to measure how well each model performed, Metrics including throughput, latency, and spectrum efficiency considered for evaluation. More accurate predictions in the rapidly changing conditions of 5G networks suggested by XGBoost, achieving a noticeably higher R2 value and showed the strongest among all the models. In 5G systems, that machine learning can help enhance resource management indicated by this study’s results. They give support towards realizing more efficient and adaptive management of networks by facilitating smarter and more flexible decisions. Overall, this study contributes to development solutions for next-generation mobile communication technologies based on AI and analytics-based. M.SC. in ETE 2025-12-21T00:00:00Z http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17437 Efficient Crawler to Enhance Dark Web Investigation for Cybercrime Detection Sarkar, Richard Jayson Historically, the Dark Web has been an online area where cybercrime has grown. This area serves mainly as a secret or hidden marketplace for reportedly goods and services exchange secretly to avoid identification. Among other operations want to sell harmful items, information such as stolen, secret information. Due to the absence of programs paired with CAPTCHA security, data encryption and the complexity of the method, technology like Tor, I2P, or Freenet, and dynamic URLs make the WebCrawler completely useless for identification and investigation. This work will suggest designing and implementing an effective dark web crawler for adaptation for use in cybercriminal activities and for qualified professional forensic analysis. During this work, many varieties of elementary data processing methods were realized from seed detection, information base clustering, machine learning categorization and structural storage to protect from crawling to facilitate forensic analysis. The feature of adaptive algorithms was acknowledged and fundamentally inflexible to access and cross the barriers. Besides, it is significantly cross-network compatible which shows the potential for sufficient coverage of services for maintain privacy. From the point of view of this experimental verification, the crawler can be significantly more accurate for the detection method and certainly meaningless regarding the amount in relation to the standard method and approaches. This research breakthrough brings in real-time investigative tools that can be both precise and cost-effective, while keeping the data undoubted forensically. In a country like Bangladesh, for instance, these devices will prove an especially strong boon. Because of these results, it’s clear that crawlers do not have to operate alone. It also has ample capacity to aid law enforcement, academicians, and big business in alleviating the swiftly growing threat of cyber criminality by boosting digital resilience. Thesis Report 2025-11-24T00:00:00Z