Authors - Britt Kristoff B. Montalvo, Vicente Pitogo Abstract - This paper presents a data-driven forecasting and anomaly detection dashboard for live births in Surigao del Norte, utilizing the Family Health Service Information System (FHSIS) data from 2021 and onwards. The research methodology is based on the CRISP-DM framework, with business under-standing for the needs of maternal services planning in the provinces and municipalities, data preparation for municipalities by quarters, time aware modeling, evaluation, and deployment through the API and visualization layer. The research employs several machine learning techniques for forecasting, such as ARIMA/SARIMA, Exponential Smoothing (ETS and Holt-Winters), and the Prophet method, along with a naïve method. The performance of the models is evaluated through the symmetric Mean Absolute Percentage Error (sMAPE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Scaled Error (MASE). A strict evaluation criterion for the deployment of the model is also implemented, such as the availability of sufficient data points in the past for the model to be deployed (i.e., 12 data points in the past), the accuracy of the model (sMAPE < 20%), and the performance of the model in comparison with the naïve method (MASE < 1). A low confidence filter is also implemented for the series with intermittent data to prevent incorrect results. The results show high reliability of the forecasting model for the entire province and better interpretability for strategic planning. However, the results also show that some of the municipalities with low population volumes and intermittent data points pose a challenge in the operation of the model.
Authors - Shriram Dange, Namdeo. M. Sawant, Sumeet S Ingole, Somnath A. Zambare Abstract - Medical image classification is of immense importance in the context of early-stage diagnosis of various neurological diseases, including Alzheimer’s disease and brain tumours. However, it remains infeasible for conventional deep learning architectures to efficiently encode frequency domain information and long-range spatial dependencies found in medical images. In this paper, a novel Hybrid Wavelet CNN Vision Trans-former, coupled with Explainable Artificial Intelligence, has been proposed for efficient and accurate medical image classification. In the proposed architecture, the application of discrete wavelet transform, convolutional neural networks, and Vision transformers for medical image classification has been presented. Additionally, explainability aspects have been addressed using the Grad-CAM technique. The proposed model was experimented with using two datasets: one for Alzheimer’s disease MRI and another for brain tumours. The experimental results reveal that the proposed deep learning architecture achieves an accuracy of 96.8%, precision of 0.96, and recall of 0.97, F1score of 0.97 for the brain tumours dataset, which beats conventional CNN, vision Transformer, and Wavelet CNN architectures. The integration of explainable AI further enhances model transparency and clinical reliability, making the proposed framework suitable for real-world medical diagnostic applications.
Authors - Sherly K.K., Merin Jose, Aleena Gerard Nidhiry, Amit Shibu Kadambamoodan, Alfahad Shahi Abstract - This paper introduces an AI-based investment assistant that helps users to understand the fundamental principles of the financial markets. This work is mainly focused on stock market data to provide accurate insights and helps in various decision-making purposes. The rising volatility in the financial markets, massive data set, and the complexity of financial instruments, makes decision-making in financial sectors more difficult to individual investors.In order to cope with this problem, our model integrates time series forecasts, large language model intelligence with real-time financial information with interactive visualizations and personalized insights. The suggested system will interpret user queries in natural language with the help of a Large Language Model (Gemini 2.5 Flash) and extracts the corresponding stock tickers and financial objects and transforms them into structured inputs to be used in predictive analysis. Past and current stock market data are retrieved with the help of yfinance API and fed into an LSTM-based time-series predictive model that predicts future price fluctuations.The results predicted are presented in interactive charts created with Plotly, which users can analyze trends easily and compare several stocks. The system can also give personalized recommendations, textual summaries of stock movements (moving up or down), multi-turn chatbot conversations, portfolio, wishlist and real time price moves besides forecasting. The proposed investment assistant improves the gap between complicated financial information and practical results by incorporating natural language comprehension, deep learningbased prediction, and intuitive visualization etc. The system promotes user knowledge and helps them in effective decision making .
Authors - Ekleen Kaur Abstract - Traditional risk frameworks, including the Geometric Brownian Motion (GBM) and stationary GARCH models, fail to account for the "volatility bursts" and "flash crashes" endemic to the altcoin market. This study the third in a series on cryptoeconomic risk introduces a multi-asset Merton Jump-Diffusion (MJD) model integrated with an Exponentially Weighted Moving Average (EWMA) covariance matrix to model portfolio risk in altcoin-only environments. By focusing exclusively on high-beta altcoins (XRP, SOL, ADA) and we address a critical gap by excluding market-anchor assets to isolate long-tail volatility dynamics neglected in existing literature. We implement a dual-model approach: a baseline MJD simulation and a "Capped Return" MJD model designed to mitigate unrealistic exponential price paths in long-horizon forecasts. Our results using Monte Carlo Value-at-Risk simulations demonstrate that incorporating a Poisson-driven jump component (j = 2.0) significantly improves λthe capture of tail risk compared to continuous models indicating pathological exponential growth without suppressing crash dynamics. Our work provides a technically rigorous framework for managing portfolios in decentralized, high-liquidity-shock environments. Backtesting via Kupiec’s Proportion of Failures test indicates that jump-based, non-stationary models achieve statistically consistent risk coverage. These findings suggest discontinuous modeling as a prerequisite for regulatory-grade risk estimation in high-beta crypto assets.
Authors - Makrand Dhanokar, Anirban Sarkar, Prajakta Dange Sant, Shivakarthik S, Krishnanjan Bhattacharjee, Swati Mehta Abstract - Named Entity Recognition (NER) is an essential task for sequence labelling and information extraction that plays a fundamental role in subsequent Natural Language Processing (NLP) applications, such as information retrieval, question answering, knowledge graph development, and machine translation. Although significant advancements have been made in NER for high resource languages, achieving effective entity recognition in Indian languages continues to be an unresolved research challenge because of linguistic diversity, complex morphology, typological differences, flexible word order, script differences, and prevalent codemixing. The scarce presence of annotated datasets and the lack of standardized evaluation metrics further limit supervised and transfer learning methods in these low resource environments. This document introduces a multilingual NER framework rooted in Sentence embeddings derived from Large Language Models (LLMs) and inference guided by prompts. The suggested method employs contextual; language independent embeddings obtained from pretrained multilingual LLMs to encode semantic representations of Indian and foreign languages within a common embedding space. Rather than using traditional token level classification, entity recognition and classification are achieved via structured prompting, allowing for zero-shot and few-shot generalization without the need for task specific finetuning. The system guarantees that entity identification and retrieval take place in the same language as the input text, maintaining linguistic accuracy and reducing error propagation caused by translation. To tackle domain variability and informal writing, constraints/guardrails for prompts and simple rule-based normalization are utilized to manage orthographic differences, script inconsistencies, and codemixed phrases often found in user generated content and social media. Experimental assessment across various Indian languages shows reliable enhancements in precision, recall, and F1score compared to traditional neural and transformer-based benchmarks, especially in low resource conditions. The findings suggest that embeddings powered by LLMs along with prompt-based reasoning provide a scalable and data efficient option for multilingual NER. This project advances the development of resilient, inclusive, and language adaptive systems for extracting information in linguistically varied settings.
Authors - Asmaul Hosna Sadika, M. M. Musharaf Hussain, Mohammad Shamsul Arefin Abstract - Credit card transaction analysis is challenged by severe class imbalance with evolving spending behavior and large-scale financial data. Many existing fraud detection approaches rely on supervised learning and assume stable fraud labels, limiting robustness under changing fraud prevalence. This study presents a large-scale, multi-year credit card trans action dataset stored in partitioned Parquet format and conducts a systematic comparison of classical machine learning, supervised deep learning, and unsupervised deep learning models for customer spend ing behavior analysis. An exploratory behavioral analysis characterizes spending heterogeneity, temporal regularities, and channel and category variations. Supervised sequence models based on LSTM and CNN ar chitectures are evaluated alongside unsupervised sequence autoencoders and hybrid detection pipelines across fraud rates ranging from 2-12%. To ensure fair evaluation under extreme imbalance, models are assessed using ranking-based metrics under fixed alert budgets, including pre cision–recall area under the curve and recall-at-K. A hybrid of Autoen coder and LSTM architectures achieves the highest performance for large systems. An integrated XAI module is introduced to derive important features providing interpretable insights.
Authors - Christian Vera, Christian Torres-Moran Abstract - This study examines how students distribute visual attention and coordinate gaze with response selection when solving image-supported multiple-choice questions in a Google Forms interface. Twenty-five students participated, selected through convenience sampling under explicit inclusion and exclusion criteria, while both fixations and click events were recorded. Oculomotor signals were processed using clustering algorithms to derive participant-specific gaze AOIs and click AOIs, complemented by a 3×3 grid-based spatial analysis to quantify global space utilization. Metrics were computed including time to first fixation, total fixation duration and fixation counts per area, transitions between areas, and the proportion of pre-response fixations within the region where the click was executed. Results show a systematic concentration of fixations in the central band of the interface, where the image and response options are located, with one or two dominant areas accounting for most fixation time. The optimal number of gaze clusters ranged from two to eight across participants, reflecting more focused versus more exploratory strategies. A high level of attention–action coupling was observed, with 80% to 95% of clicks occurring within the same area that concentrated most fixations. These findings support the use of eye track-ing as a tool for cognitive validation of item design and inform principles for more efficient and transparent digital assessments.
Authors - V. Abarna, R. Shyamala Abstract - The rapid advancement of artificial intelligence has significantly enhanced deepfake generation techniques, posing serious challenges to digital media authenticity, cybersecurity, and misinformation control. Conventional detection approaches often rely on single-modality analysis, limiting their effective-ness against sophisticated synthetic media. This paper proposes a multimodal deepfake detection framework that integrates visual, audio, textual, and behavioral biometric information using a hybrid deep learning architecture combined with a variational quantum learning approach. Deep neural models are employed for feature extraction across modalities, including convolutional networks for visual artifacts, transformer-based models for speech and text analysis, and bio-metric behavioral assessment such as eye movement, lip synchronization, and motion consistency. A hierarchical fusion mechanism aggregates modality-specific representations, while a variational quantum classifier enhances classification robustness through hybrid quantum–classical learning. An explainability module provides insight into modality contributions and prediction confidence, supported by a web-based dashboard for real-time interaction. The proposed framework aims to improve detection reliability, interpretability, and practical deployment in applications such as digital forensics, social media verification, and cybersecurity. This work presents a conceptual architecture and implementation roadmap to support future research in multimodal deepfake detection.
Authors - Emine YAZICI, Alper UGUR Abstract - Critical infrastructures are of strategic importance to the security of societies, economic stability, and the continuity of public services. However, with digitalization, these infrastructures are facing progressively complex cyber threats such as supply chain exploitation, ransomware, and AI-assisted targeted attacks. Traditional hardening methods are becoming insufficient in the face of these developments. This study examines the types of attacks and threat trends that have emerged in the literature in recent years; and evaluates the effectiveness of hardening methods applied against them at the software, physical, and organ izational levels. The findings indicate that, due to the dynamic nature of threat vectors, utilized common risk analysis and hardening strategies are insufficient to deliver the expected security outcomes. However, the literature lacks a risk analysis score and hardening guide for decision-makers regarding current threat models and attack techniques. In this study, risk scores based on CVSS were cre ated for up-to-date threats in the ICS field, and hardening mechanisms were also proposed according to the mechanisms behind the related threats and their ef fects. We aim to address existing shortcomings to some extent by calculating the risk scores of new attacks and to make ICS more secure through proposed hard ening mechanisms against these risks. The sustainability of security can be achieved through holistic security policies that include multi-layered approaches, continuous monitoring, adaptive response mechanisms and advanced approaches such as Zero Trust architecture, AI-based anomaly detection, and hybrid defense systems in the domain where traditional measures fall short.
Authors - Nurkholis, Katherin Indriawati Abstract -This paper presents a case study on a High Voltage Jet (HVJ) electric boiler, focusing on current unbalance (CU) risk identification and mitigation us ing a combined data-analytics and Failure Mode and Effects Analysis (FMEA) framework. Power-quality assessment follows IEC 61000-4-30 for voltage un balance (VU), while CU interpretation refers to NEMA MG-1 and IEEE recom mendations. The proposed workflow integrates (i) instrument classification (Class A for voltage), (ii) time synchronization across logger/PLC/power-quality analyzer to avoid timestamp drift, and (iii) historian-based data pre-processing (outlier cleaning, scaling, and missing-data handling) prior to statistical analysis. Results show an average CU of 6.85% with a standard deviation of 0.48% and a maximum of 15.92%, indicating operational periods exceeding common industry limits. FMEA highlights electrode aging/damage, loose/corroded cable connec tions, and supply power-quality issues as the dominant contributors. Recom mended actions include online phase-current monitoring, improved water-chem istry and blowdown management, and control optimization of the VFD-driven boiler circulation pump (BCP).
