Authors - GGS Pradeep, Thrilok. Kolla, Rajesh Sharma R, Akey Sungheetha, N Vijayalakshmi, Pellakuri Vidyullatha Abstract - Ambient Intelligence (AmI) systems are intended to support responsive and adaptive services by understanding the context of their users. The present paper proposes a machine learning framework for context prediction in ubiquitous environments on the basis of multi-sensor ambient data. The method classifies user activity into four high-level contexts: Sleeping, Working, Cooking, and Outdoors, exploiting features like Temperature, Light, Sound, Motion, and Time. The model described here consists of three steps: Preprocessing, Normalization, and Classification based on Random Forests. Visualization techniques, such as KDE plots, correlation matrices, and 3D scatter analysis, are used for the interpretation of contextual separability and the influence of features. Experimental results show that outdoor activity strongly dominates the dataset, while other contexts are almost neglected. The proposed framework shows high predictive accuracy with interpretable insights into context-dependent behavioral patterns. This work thus contributes to the design of such adaptive, context-aware systems that can find a home in intelligent environments, especially in the realms of smart homes and assistive technologies. The abstract makes it abundantly clear how crucial Ambient Intelligence systems are to contemporary ubiquitous surroundings, particularly for assistive technology and smart houses.
Authors - Adrian Diepeveen, Siphesihle Sithungu Abstract - Traditional electricity grid networks have approached critical resilience thresholds, with archaic grid architectures demonstrating inadequate capacity to accommodate volatile demand fluctuations. South Africa's current energy crisis, characterised by unpredictable power outages and extensive economic losses from electricity outages, emphasises the need for novel optimisation frameworks. This re-search proposes a framework named AMPERE (Autonomous Multi-Agent Predictive Electricity Reinforcement Learning Engine), a dual-phase solution which com-bines hierarchical multi-agent deep reinforcement learning (MADRL) and evolutionary algorithms (EAs) in order to address electricity distribution challenges through decentralised multi-agent coordination. A significant gap exists in research integrating distributed hierarchical knowledge-sharing MADRL with EAs at both regional and national levels, since current approaches focus on singular artificial intelligence (AI) methodologies or non-hierarchical frameworks. Ultimately, AMPERE introduces a novel three-tier hierarchical agent architecture consisting of smart grid national agents, smart home regional agents, and smart home battery agents, utilising Pecan Street's real-world Internet of Things (IoT) time-series datasets. The Centralised Training with Decentralised Execution (CTDE) paradigm enables effective multi-agent coordination across regional and national scales, subsequently enhancing smart grid stability and scalability while minimising electricity lost in distribution.
Authors - Jayan V, Sreejith Alathur Abstract - The COVID-19 pandemic not only strained global healthcare systems but also exposed healthcare professionals to unprecedented psychological challenges. Among the various stressors, the rapid spread of misinformation on social media emerged as a significant contributor to emotional distress, confusion, and burnout among frontline workers. This study explores the impact of social media misinformation on the mental health of healthcare professionals in India during the pandemic, using qualitative interviews with 41 public sector healthcare workers. Guided by the honeycomb social media framework and stress coping models, the re-search examines how misinformation intensifies psychological stress and how individuals and institutions respond to it. Findings reveal that misinformation created confusion in clinical protocols, fueled public panic, and intensified professional anxiety. Participants emphasized the need for verified communication channels, stronger digital literacy, and mental health support systems. The study underscores the critical role of individual responsibility, institutional action, and policy reform in combating the "infodemic" and protecting the well-being of healthcare workers. Recommendations include the establishment of centralized information platforms, community-driven myth-busting initiatives, and proactive involvement of enforcement agencies. This research calls for a multi-stakeholder approach to address misinformation as a public health risk in its own right.
Authors - John Paul C. Masuhay, Reynaldo R. Corpuz Abstract - This piece of work, setting up a new stage, introduces a novel optimization approach for Torus Fully Homomorphic Encryption (TFHE) using Genetic Algorithms (GA). TFHE protects data privacy by facilitating calculations on encrypted data. The study integrates GA with TFHE to optimize key parameters, such as LWE dimension, decomposition level, and security level, aiming to achieve an optimal trade-off between encryption accuracy, bootstrapping time, ciphertext size, and security. The GA optimization uses a multi-objective fitness function, in which performance metrics are derived from synthetic datasets such as text, alphanumeric, and special characters for simulation. The results indicate that the GA framework using DEAP was effectively integrated with TFHE. After five generations, the optimization process attained a best fitness value of 0.9691, marking significant enhancements in system performance. Specifically, the GA optimization improved TFHE parameters, such as increasing the LWE dimension from 1024 to 1514.02 by reducing the noise standard deviation from 3.2 to 2.53 and improving the security level from 128 bits to 242.57 bits. The decomposition level was adjusted from 4 to 2, optimizing computational performance while slightly reducing security. Variation in bootstrapping times across datasets reflects how efficiently the optimization balances efficiency and performance. These remarkable findings suggest that GA could potentially be applied to im-prove TFHE parameters, making it a more efficient and secure method to perform privacy-preserving computations, particularly within edge computing environments.
Authors - Alpa R. Barad, Ankit R. Bhavsar Abstract - A good quality of forage is essential to feed cattle to improve its health and productivity. Quality of grass has been affected by various factors such as weather condition, leaf disease, and diverse species with texture similarity. Over multiple species of grass and variation over sessions leads quality recognition more complex and difficulty. Proposed study uses unsupervised approach to reduce dependence of annotation over multiple species of grass. Proposed work uses median and fast denoising algorithm to enhance input image. Article present novel approach with integration of Gabor filter with k-means++ model. Gabor filters helps to identify optimized features over texture similarity problem of grass. Simulation of proposed study also measures denoising input grass image with 0.72 SSIM. Simulation of proposed study also finds remarkable performance with 79.86% accuracy with median and Gabor filter. Simulation of study also explore possible variations with hybrid unsupervised approach to enhance performance of the model.
Authors - GGS Pradeep, Thrilok. Kolla, N Vijayalakshmi, U Ananthanagu, Rajesh Sharma R Abstract - The booming volume of genomic data requires computational techniques that will allow the efficient extraction of biologically meaningful patterns. This work investigates advanced data mining algorithms for extracting frequent patterns and motifs in DNA and RNA sequences based on k-mer analysis. The nucleotide sequences are divided into shorter pieces called k-mers, where frequent pattern mining, positional frequency mapping, and dimensionality reduction techniques are applied to expose conserved motifs and structural characteristics. The proposed methodology employs Apriori for mining, followed by Principal Component Analysis (PCA) and a variation of a sequence logo style for visualization, which exposes latent relationships and biologically relevant sequence patterns. The experimental data show uniformly distributed occurrences of dominant k-mers and positionally clustered functional motifs like the “ATG” codon. The PCA projections in 2D and 3D space further highlight the structural diversity and possible clustering of sequences according to k-mer profiles. This research extends the frontier of bioinformatics by aiding in the interpretation of genomic sequences, with applications in genome annotation, disease-associated gene prediction, and regulatory motif identification. It serves the purpose of informing about the methods being brought together under k-mer analysis, frequent pattern mining, PCA, and motif visualization.
Authors - Zlatinka Kovacheva, Kalinka Kaloyanova, Ina Naydenova, Mariana Trifonova Abstract - This paper concerns project-based learning. The aim of the study is to compare the results of group and individual projects of undergraduate students studying Information systems. The research is based on statistical inference and hypotheses tests. A hypothesis test is presented regarding a difference between means of two dependent samples for quantitative indicators that have a normal distribution. The result analysis is based on 3-years experiments with student pro-jects. For all experiments, the null hypothesis is rejected, and in both methods of assessment (group and individual projects), different results are obtained. For the majority of the groups involved in the experiments, the marks of the group project are higher than the average marks of the individual project but there are also exceptions. The results show that students who performed better on the individual project often received lower grades on the group project. Also, students with low individual grades received higher scores on the group assessment. As a conclusion, we can recommend applying both methods – group and individual projects for students’ assessment, paying attention at their advantages and disadvantages.
Authors - Nguyen Hoang Kha, Su Truong Phuc, Huynh Tu Anh, Ha Cao Vi, Tran Quoc Hao, Vinh Dinh Nguyen Abstract - Face recognition tasks often suffer from poor performance when limited real training data is available per class. To address this, we propose a novel pipeline that combines LoRA-based fine-tuning of a generative model with synthetic image generation and YOLOv8 classifier training. Our method fine-tunes Stable Diffusion using just a few real face images per identity, then generates realistic, identity-preserving synthetic images to augment the training dataset. The augmented dataset is then used to train a YOLOv8n classifier for face recognition. We evaluate our pipeline on a 31-class face dataset and achieve a Top-1 classification accuracy of 80.05%, significantly outperforming the baseline YOLOv8n model trained only on resized real images (60.94%). These results demonstrate the effectiveness of our method in low-resource settings for face recognition.
Authors - Keshav Dudani, Nischay Jain, Rishi Raj, Nitesh Patnaik, Vishal Meena Abstract - Stock market volatility and the effect of financial trends and external news perspectives are reasons why predicting the direction of the stock market is so intrinsically tough. Conventional models may fail to provide accurate forecasts when they fail to react to the quick changes in the marketplace. Our proposed method boosts the success of predicting by employing StockVisionX, which constructs classifications by combining past stock movements and sentiment analysis of financial news sources. The sentiment polarity that is integrated into StockVisionX delivers a more thorough market movement projection, thereby forecasting whether a stock will gain or drop in price. StockVisionX adds even more strength in its predictability with the usage of vital technical indicators such as SMA, EMA, RSI, MACD, ATR, and OBV. Supplementing these criteria with sentiment analysis ensures a better and evidence-based manner of categorizing stock movement. The model is trained and tested on diverse datasets to analyze sentiment and forecast prices, and task-specific optimization is seen. Our Naive Bayes technique is 54.5% more accurate when compared to Vader sentiment categorization. Our Prophet + XGBoost model, which has been trained on non-correlated real-world data, outsmarts ARIMA and LSTM by 49.0% and 29.6%, respectively, when predicting price; however, it is hard to determine if these models would work better with highly correlated data. This is evidence of the power and generalizability of our process. StockVisionX is efficient and precise in its classification model on predicting stock movement through the meticulous integrating of sentiment-related information with the trend history.
Authors - GGS Pradeep, Thrilok. Kolla, Rajesh Sharma R, Akey Sungheetha, N Vijayalakshmi, Pellakuri Vidyullatha Abstract - The ever-increasing implementation of IoT sensor networks in smart environments has created a multitude of multivariate spatiotemporal data streams. The effective identification of temporal patterns and anomalies in those streams becomes necessary to ensure environmental awareness and operational integrity. This paper proposes an elaborate framework for temporal pattern recognition in IoT sensor data through the integration of spatiotemporal reasoning, statistical modeling, and anomaly detection. The approach uses statistical descriptors, kernel density estimation, correlation matrices, and spatiotemporal visualization to capture local anomalies as well as trends prevalent in aggregation over distributed sensors in order to illustrate its strengths in analyzing the data. One of the case studies depicts the environmental awareness based on monitoring temperature: deviation detection, inter-sensor coherence, and a conceptual view of anomaly propagation. The results indicate the potential of establishing a system capable of real-time and interpretable monitoring in dynamically evolving IoT settings. This work contributes toward more adaptive and transparent sensing architectures capable of operating under uncertainty and environmental variability. The abstract is coherently structured, flowing from motivation to case study, methodology, results, and contribution. It also accurately frames the rationale for the study and successfully defines the fundamental difficulty in IoT sensor networks, which is the recognition of temporal patterns and abnormalities.
