MKSSS's Cummins College of Engineering for Women (CCOEW) is an Autonomous engineering college in Pune, Maharashtra, India established in 1991 and run by the Maharshi Karve Stree Shikshan Samstha.
Stillbirth is one of the biggest causes of newborn deaths globally. Noninvasive predictive approaches involving electrohysterography (EHG) have been extensively researched. Nonetheless, most EHG-based predictive models consider recordings as independently collected samples and emphasize prediction between spontaneous pre-term and term births without taking into consideration those that end up in induced and cesarean births. The newly introduced Induced Cesarean EHG Dataset (ICEHG DS) offers a distinct chance to explore the temporal behavior of EHG recordings during pregnancy. In this paper, we present a framework for capturing the temporal behavior of EHG recordings through calculating temporal, spectral, and nonlinear feature differences between recordings at an earlier (23-weeks gestation) and a later (31-weeks gestation) stages of the same pregnancy. Eighteen features from each recording representing temporal, spectral, and nonlinear domains are extracted and the difference between the two points in time are calculated. A multi-classifier evaluation framework-comprising neural networks, random forests, and support vector machines-is employed to predict delivery outcomes (induced, cesarean, or induced-cesarean). Our results demonstrate that temporal delta features, particularly frequency-domain and nonlinear measures, provide discriminative information for delivery mode prediction. The MLP classifier achieved the highest test-set accuracy (66.7%), while the Random Forest classifier showed more stable cross-validated performance, with frequency-domain delta features emerging as the most important predictors. To the best of our knowledge, this is the first study to systematically evaluate temporal delta features on the ICEHG dataset, establishing a foundation for longitudinal EHG analysis and highlighting the potential of tracking feature evolution as a clinically meaningful biomarker.
Financial fraud is become one of the biggest challenges for banks, online payment platforms, and insurance firms. It leads to financial losses overall stability of the financial system. To deal with this researchers are now using advanced methods such as machine learning (ML) and deep learning (DL). These approaches can study very large amounts of data, uncover hidden trends, and adjust more quickly to new fraud behaviours. This paper reviews studies published. Each focusing on how different ML and DL models are used to detect fraud. Deep learning models like Long Short Term Memory (LSTM) and Gated Recurrent Units (GRU) work especially well on transaction sequences, where the order and timing of payments matter. On the other hand, classical ML models such as Random Forest, XGBoost, and K-Nearest Neighbours (KNN) are still widely used because they are simpler to explain, faster to train, and perform strongly in many practical fraud detection cases. In short machine learning and deep learning helps us in making these systems balance speed and accuracy
In high-hazard workplaces like packaging facilities, effective fire safety is critical, but conventional practices fail to recognize actual hazards and are highly expensive to implement. This paper presents a hybrid reconstruction and artificial intelligence-driven framework that can potentially be applied to build interactive virtual reality environments. The objective of this study is to develop a scalable and cost-effective Virtual Reality based fire safety training system that balances realism and interactivity. To balance visual fidelity and interactivity, a hybrid reconstruction pipeline was developed. The complex background environment was reconstructed and rendered using 3D Gaussian Splatting, while for reconstructing key industrial objects as solid and interactive meshes, photogrammetry is used. An artificial intelligence-based system has been adopted for automatic object detection using You Only Look Once version 11 (YOLOv11) and material-based hazard classification using Bidirectional Encoder Representations from Transformers (BERT). In addition, interaction options are generated using a text generation model Fine-tuned Language Net Text-to-Text Transfer Transformer (FLAN-T5). The results indicate that the proposed framework produces high rendering capabilities with high precision, enabling efficient and scalable development of industrial safety training modules.
Query-driven video summarization creates concise video summaries tailored to the user’s textual queries. Unlike traditional summarization methods that produce generic video summaries, query-driven approaches focus on the relevance of the summarized content to the user’s specific interests. These methods leverage Natural Language Processing and Computer Vision techniques to analyze and understand both the video and the query, enabling the extraction of keyshots relevant to the user’s interest. This research introduces a methodology for query-driven video summarization via shot-level caption generation. The method involves segmenting the input video and generating shot-level captions using a pretrained Bootstrapped Language-Image Pre-training (BLIP) model. Both the generated captions and the user-inputted textual queries are encoded into token-level features using a Contrastive Language-Image Pre-Training (CLIP) text encoder. Contextual mapping is performed to compare the captions with the input queries, and matching keyshots are extracted. The performance of the proposed method is evaluated using F1-scores. The qualitative and quantitative experimental results demonstrate that the proposed technique successfully produces query-driven video summaries.
This paper presents a real-time, vision-based system for Indian Sign Language (ISL) recognition and translation, aimed at enhancing communication between the deaf community and non-signers. The system combines a CNN-LSTM architecture for static gesture recognition, achieving an accuracy of 98.47
