Sardar Patel Institute of Technology (SPIT) is an autonomous un-aided Research and Engineering Institute affiliated to University of Mumbai. The college was established in 1995 as an extension of its sister institute, the Sardar Patel College of Engineering, before becoming an independent unaided institute in 2005. S.P.I.T is one of the premier technical institutions of Maharashtra, offering undergraduate, postgraduate, and doctoral programs in engineering and computer applications. Dr. Bhalchandra Chaudhari is the current principal of the institution.P.P.I.
The advancements of technology in every domain are leading to the misuse of data, especially in information sharing. Many ways have been introduced for information sharing in the digital world, and digital images act as a powerful way to share information. With the development of image editing software and the availability of mobile devices, the manipulation of images is very easy, and it is challenging to identify the manipulated images. People mainly use forged images to spread rumours and can damage the public trust in media; hence, it is necessary to develop an effective forgery detection task. Various researchers, image forensic experts, have been working on image forgery detection and have identified some issues, including high computational complexity, inaccurate classification, and a lack of depth feature analysis. So the proposed approach is intended to develop a novel physics based forgery detection approach with efficient feature extraction. The digital image’s shadow features, handcrafted and deep features, were extracted using Hough Transform, Sobel Operator, and ResNet-151 module, respectively. These extracted multi-features are fused using a novel Squeeze Excitation based Vision Transformer Network (SE-ViTN) for better feature embedding representations. Furthermore, the embedded feature patches are classified as normal or forged using an Optimized Radial Basis Support Vector Machine (ORB-SVM) classifier, and their parameters are tuned using the Secretary Bird Optimization Algorithm (SBOA), which reduces the complexity and enhances efficiency. This research is validated on two benchmark datasets, namely DSO-1 and the OIS dataset. Further, experimental results showed better performance accuracy of 98.83 and 97.75
In high-stakes emergency response scenarios, the ability to automatically detect urgency from audio inputs can significantly improve response times and save lives. However, most existing systems focus primarily on emotion recognition or rely on clean, structured datasets, which do not adequately capture the chaotic and stressful nature of urgent real-world communication. This research introduces an innovative method for urgency classification using speech features derived from 911 emergency calls and public datasets such as RAVDESS and LibriSpeech. A carefully balanced and realistic dataset was developed, with noise augmentations to better simulate the acoustic diversity found in real-life scenarios. Baseline performance was established using conventional machine learning algorithms. To improve model interpretability and feature efficiency, SHapley Additive exPlanations (SHAP) has been integrated to guide feature selection via an elbow-based importance cut-off. Performance was assessed using common evaluation metrics, including accuracy, precision, recall, and F1-score. These findings demonstrate that XAI-driven feature selection can simplify urgency detection models without sacrificing predictive performance, offering a promising foundation for real-time emergency response systems.
Affective assessment is emphasized in National Education Policy (NEP) 2020 for holistic education but scalable measurement methods to implement it are currently limited in educational institutes. A review of physiological, linguistic, behavioral, and multimodal approaches for automatic personality recognition (APR) has been done in this paper to identify methodological trends and practical deployment limitations. Common challenges identified across all approaches include lack of scalability, inadequate psychometric grounding, limited crosscontext generalization, and dependence on expensive hardware. Based on these findings, a conceptual end-to-end framework is proposed in this paper in which behavioral telemetry from routine computer-based assessments is utilized to infer Big Five personality traits. The suggested framework combines supervised and sequential modeling, psychometric benchmarking, and explainable AI, providing a blueprint for non-intrusive and scalable affective assessment. The empirical implementation of proposed framework is reserved for future work.
Traditional automated grading systems often rely on shallow linguistic features and basic machine learning models, struggling to capture deeper semantic and contextual nuances. Manual assessment, while nuanced, is increasingly unsustainable due to high teacher-student ratios, resulting in inconsistent and time-intensive grading.This research introduces an advanced Automated Grading System (AGS) with dual-component architecture. The first component integrates Word2Vec embeddings, Latent Dirichlet Allocation (LDA), and Long Short-Term Memory (LSTM) networks, achieving a quadratic-weighted Cohen’s Kappa score of 0.9675 on the ASAP dataset. The second component leverages BERT sentence transformers for semantic analysis using cosine similarity, ensuring context-aware evaluation.Additionally, a quiz generation module powered by T5 transformers automates MCQ creation with pedagogically relevant distractors, streamlining quiz design and improving learning outcomes. The proposed AGS provides a reliable, scalable, and context-sensitive solution for modern educational environments.
