Muhammad Nazrul Islam

Deep learning fire detectors are usually optimised for in-domain accuracy, while their cross-dataset generalisation, which is vital for real-world deployment, remains poorly understood. We propose a dual-teacher knowledge distillation framework that compresses a YOLOv8m teacher (25.84M parameters, 78.7 GFLOPs) into a GhostConv-enhanced YOLOv8n student with efficient channel attention (2.298M parameters, 6.2 GFLOPs) through spatial feature mimicry and CLIP ViT-B/32 semantic guidance. The distilled student achieves 72.0\% mAP@50 on the D-Fire validation set (four-seed mean), retaining 91.6\% of teacher performance with 11.2$\times$ fewer parameters, and 71.5\% mAP@50 on the held-out test set at 375.6 FPS with TensorRT on an RTX 5080, exceeding the closest distillation baseline by 9.3 percentage points. Ablations show that CLIP semantic distillation alone reaches 72.6\% mAP@50, surpassing the combined loss and revealing the importance of the transfer mechanism under architectural mismatch. Across four external benchmarks, however, all eight tested models fall from 71--79\% mAP@50 on D-Fire to $\leq$3.3\%, and CLIP-based confounder prompts do not reduce false positives relative to detection-only training. This first multi-benchmark generalisation analysis of a sub-5M-parameter fire detector shows that distillation delivers edge-deployment gains but leaves cross-domain robustness unresolved.

The transition from student life to professional employment represents a challenging shift marked by emotional and practical adjustments. This research explores the sentiment dynamics of two surveys conducted with two different groups from the Science and Technology background: Fresh graduates who joined the workplace and students who are in their final phase of academic life and likely to start their professional life. Exploratory data analysis was performed to process numerical and categorical responses, while an autonomous thematic analysis was introduced after implementing similarity-based clustering using K-Means in open-ended responses. Subsequently, Google Pegasus-Large has been employed to summarize open responses and assign preliminary clustering to the responses, and the GPT-5.1 model has been used to extract themes and codes with sentiment labels, which were later reviewed through manual validation. By capturing and analyzing these sentiments, the research identifies patterns in expectation management, resilience building, and coping mechanisms for workplace stress. In this article, we have provided guidance for targeted employee support and organizational adaptation, supporting human factors–informed interactive tools to optimize user experiences, easing the academic-to-work transition for newprofessionals. Finally, this research provides a foundation for improving employee well-being, effective organizational practices, and healthier work environments.

Cureus Journals Clinical decision-making, in most of the cases, is convoluted, perplexing, and critical, and becomes more delicate in case of burn patients where specialized healthcare services are required. In an emergency situation, various factors are needed to be considered within a limited time bound to take pivotal decision, such as the rate of replacement of the bodily lost fluid due to burn. In such circumstances, a computer-aided intelligent clinical support system is doubtlessly useful to make intuitive decisions in a shortest possible time, where a fuzzy inference model (FIS) may lead to the solution. In this study, an improved FIS-based fluid resuscitation model is proposed for burn patients, where the rate of required fluid revitalization is determined intelligently. The goal of this study is to suggest an assistive model for the physician in making a quick decision for intravenous fluid rate (IFR) adjustment depending on the percentage of total body surface area (%TBSA) burned and hourly urine output (HUO) measurements in burn patients. To attain the objective, the proposed FIS model is introduced with the %TBSA and HUO as input variables and IFR as the output variable of the system. These two inputs are considered in this research due to their foundational role in classical calculations, as all existing formulas are derived from them. This study also conducts a comparative analysis among various defuzzification methods for various membership functions to determine the best possible model. The simulated results ensure the viability of proposed model for practical experimentation where centroid of area with Gaussian membership function is recommended. Furthermore, the problems of traditional estimation of IFR such as discrete rate of fluid injection and abrupt shift (≈ ±20%) of IFR level depending on the variations of HUO would be resolved by the proposed FIS model, resulting in a smooth and continuous adjustment of IFR. Nevertheless, the proposed design shows significant improvements in determining IFR, as validated through benchmarking.

Road traffic accidents present a major public health and economic challenge in Bangladesh, with thousands of lives lost annually due to poor road infrastructure, human error, and lack of effective safety measures. This study proposes a comprehensive machine learning-based framework for predicting accident intensity using real-world road crash data from 2007 to 2024, comprising 46,256 records obtained from four national sources. The dataset was preprocessed through imputation, feature engineering, and label encoding, followed by the application of four parallel modeling pipelines: traditional machine learning, ensemble learning, PyCaret-based automated modeling, and neural networks. Among all models, the Extra Trees classifier achieved the highest performance with 96% accuracy, outperforming existing approaches in both prediction and reliability. To enhance model transparency, explainable AI techniques such as SHAP and LIME were applied, revealing ’Location’, ’Month’, and ’Weather’ as the most influential features. The findings not only offer robust predictive insights but also support data-driven interventions for policymakers to identify high-risk zones and design targeted road safety strategies.