Authors - Palash Sontakke, Pankaj Chandre, Prashant Dhotre, Ganesh Pathak Abstract - The dyeing segment of textile manufacturing faces pressing demands to curb water, energy and chemical consumption while maintaining colour con-sistency and fast throughput. This paper presents a layered edge‑to‑cloud archi-tecture that employs artificial‑intelligence techniques to manage resources dy-namically across the entire dyeing supply chain. IoT sensors on each dyeing machine stream high‑frequency telemetry to an AI cloud where data are curated in a scalable lake and enriched through feature engineering. Deep‑learning models forecast short‑term demand, yields and process drifts, enabling proac-tive recipe adjustments. A digital‑twin simulation engine mirrors every dye bath and tank, permitting insilico experimentation of eco‑friendly formulas without risking live production. Reinforcement‑learning agents, trained against the twin, generate near‑optimal job schedules and set‑points that minimise water and chemical footprints while respecting delivery constraints. All recommendations pass through an explainable‑AI module that surfaces the key factors driving each decision to foster operator trust and regulatory compliance. A blockchain ledger records batch‑level resource transactions, ensuring tamper‑proof tracea-bility for audits and sustainability certifications. The architecture integrates seamlessly with existing ERP/MES systems, closing the loop between predic-tive intelligence and shop‑floor actuation. Proof‑of‑concept results from a mid‑scale dye house show up to 18 % water savings and 12 % energy reduction, with payback achieved in under nine months. Collectively, the proposed framework demonstrates how advanced AI and Industry 4.0 technologies can deliver both economic and environmental gains in dyeing operations.
Authors - Princessa Natasya Kwang, Yanthi Hutagaol-Martowidjojo Abstract - This research analyzes the effect of ESG disclosure scores on stock price synchronicity among Australian mining companies, exploring each pillar's contribution. The final sample consists of 275 firm-year observations of Australian listed mining companies covering the period from 2020 to 2024. Data was collected from Bloomberg and Factset. This study adopts the price synchronicity model proposed by Hutagaol-Martowidjojo et al. (2023) and panel data regression is employed to analyse and test the research hypotheses. The study concludes that ESG disclosure is significantly associated with higher price synchronicity, implying that while ESG disclosure scores enhance overall market visibility, they may simultaneously reduce the uniqueness of firm-specific information.
Authors - Thejan Dulara, Indra Mahakalanda, Prasanga Jayathunga Abstract - Media budget allocation remains a major challenge in the advertising industry. Inefficient spending and biased budgeting decisions often cause un-derperforming campaigns. It is a challenge for advertisers to strike the right balance between traditional media like TV, radio, press and the digital media. This not only wastes resources but reduces campaign impact. Media mix problem revolves around media performance and return on investment. This study develops a data-driven media mix determination model using machine learning and genetic algorithms. The goal is to maximize audience reach while minimizing costs. The model focuses on key quantitative factors such as cost-effectiveness, media efficiency, and saturation points. The use of data collected from the Sri Lankan market ensures the practical relevance of the study. Supervised machine learning models such as decision trees, random forests, XGBoost, and LightGBM are tested to understand the complex, non-linear behavior of media performance. The above models predict how different channels respond to increased spending. Curve smoothing enables the identification of saturation point and efficiency levels. A genetic algorithm deployed to identify the optimal budget allocation across media platforms. The result provides a practical media mix for the Sri Lankan advertising industry where advertising planners can allocate budgets more effectively with minimum overspends to achieve higher campaign success. Although the models were learned using a country specific data set, they can be used for similar scale markets. This study contributes to the existing literature by developing an efficient method for me-dia advertising planning.
Authors - Thejan Dulara, Indra Mahakalanda, Prasanga Jayathunga Abstract - Efficient media scheduling plays a vital role in advertising campaign success. Among media platforms, television remains a key channel for audience engagement. However, allocating advertisements across programs, times, and channels is a complex task. The number of possible combinations exceeds human evaluation capacity. Advertisers often face challenges in allocating spots effectively while minimizing plan cost. This study implements a Mixed Integer Linear Programming (MILP) model to optimize spot allocation across TV programs. The objective is to minimize the Cost Per Rating Point (CPRP), a key efficiency metric. The model maximizes the normalized gross rating point (GRP), subject to constraints such as total budget, program slot availability, and fair commercial distribution. By transforming a nonlinear objective into a solvable MILP framework, the model ensures practical feasibility and scalability. The results demonstrate improved CPRP performance and significant time savings under realistic constraints. This approach supports advertisers in making data-driven, cost-effective media scheduling decisions. The study offers a replicable framework for optimizing television advertising plans in dynamic markets.
Authors - Aishwarya Anand Arora, Charu Gupta Abstract - The explosive growth of artificial intelligence (AI), particularly generative AI, poses opportunities and challenges gender equality. AI has the capacity to shatter cultural boundaries and build inclusive systems, yet latent algorithmic bias and gender inequality in its composition can exacerbate existing divisions. This overview delves into the ways generative AI can both perpetuate or decrease gender bias, contingent upon how responsibly it is created and deployed. AI systems tend to reflect the biases of the society they are designed from. For instance, AI recruitment tools have tended to select more male candidates because of gender-biased historical data. Generative AI, too, tends to underrepresent women in leadership positions and perpetuate stereotypes. If unregulated, these systems stand a chance of perpetuating gender imbalance across industries. In this paper, we have discussed how AI also has the potential to advance fairness. Resume-anonymizing tools, huge data set analysis to detect biased tendencies, and individualized learning plans can promote gender inclusiveness. AI can help bridge STEM access disparities for women and provide supportive mentoring networks in education. To harness these benefits, responsible AI governance is critical. Fairness audits, diverse datasets, transparency, and accountability should guide AI development. Increasing women’s participation in AI research and policymaking is also key to creating more equitable systems. Ethical AI guidelines must ensure that AI use is inclusive, fair, and non-discriminatory. Generative AI has the potential to be an effective catalyst of gender equality when coupled with ethical governance, representative data, and representation diversity, leading towards a more equitable, AI-powered world.
Authors - Pujari Suresh Kumar, Sushama Rani Dutta Abstract - Image Inpainting is the field that reconstructs the missing regions with high-quality image restoration. The image restoration addresses the loss of conditional details and distortion of visual quality in damaged images. In this paper, a Scale-Variant Learning (SVL) model is presented that combines super-resolution and deep image inpainting for the restoration of missing or damaged regions, resulting in increased image clarity and high resolution. The proposed SVL consists of a super-resolution network that first recovers high-frequency details from low-quality input and an inpainting module that fills in missing re-gions by leveraging both contextual and structural information. To generate photo-realistic output, a multi-scale feature fusion, attention mechanisms, and adversarial learning are used to enable the model learns the global semantics and local texture consistency. The pre-trained model DeepFill v1, with transfer learning applied to large-scale high-resolution datasets (CelebA-HQ and Plac-es2), shows competitive performance compared to existing techniques in both quantitative evaluations and visual appearance. This pre-trained model fills the gap between missing content and accurate regions, from image restoration and editing to historical image repair and computer vision tasks that require fine-grained representation.
Authors - Tapiwa Gundu, Nangamso Mmango Abstract - Small and Medium Enterprises (SMEs) in South Africa face signifi-cant cybersecurity challenges due to poor password hygiene practices. Employees and managers frequently experience password fatigue, leading to dangerous security shortcuts such as password reuse, weak credential creation, and physical recording of passwords. While comprehensive cybersecurity frameworks exist for large enterprises, SMEs lack practical, tailored solutions that account for their limited resources and technical expertise. This study conducts a systematic literature review (SLR) to analyze existing research on password security challenges in SMEs, with particular focus on the South African context. The review identifies key vulnerabilities and synthesizes evidence-based recommendations into actionable password hygiene guidelines specifically designed for SME environ-ments. The proposed guidelines emphasize usability alongside security, offering realistic strategies to improve authentication practices without overwhelming staff. Findings highlight the critical need for simplified policies, appropriate tech-nological tools, and continuous security awareness training in resource-constrained SME settings.
Authors - Shruthi S, Manjula Y Abstract - Automated emotion detection through facial analysis represents a cru-cial technological advancement for interpreting human expressions from digital images. Current classification approaches face challenges in achieving precise accuracy because they extract insufficient distinguishing characteristics needed to differentiate various emotional states. To address these issues, we introduce a Parallel Deep Convolutional Neural Network (PDCNN) incorporating Gaussian Error Linear Unit (GELU) activation for enhanced emotion classification performance. Our methodology utilizes two established benchmark collections: FER-2013 and Japanese Female Facial Expression (JAFFE) databases, which undergo preprocessing through Contrast Limited Adaptive Histogram Equalization (CLAHE) and dataset balancing via Generative Adversarial Network (GAN) based augmentation techniques. Subsequently, the enhanced dataset undergoes feature extraction using DenseNet-121 architecture, which captures representa-tions through densely connected layers before processing through our dual-stage classification framework. This system leverages both localized and comprehen-sive facial characteristics, integrating them via a fusion mechanism for precise emotional categorization. Our experimental validation demonstrates that the PDCNN with GELU approach achieves superior performance with 94.67% ac-curacy on FER-2013 and 99.10% on JAFFE datasets, outperforming traditional CNN, Convolutional Relational Network (CRN), ResNet-50, Wavelet DCNN, and Three Channel-CNN methodologies.
Authors - Sneha Warrier, Umesh Mahtani, Smitha Rao Abstract - This study makes a novel attempt using ensemble Machine Learning (ML) techniques to analyze the impact of various Environmental, Social, and Governance (ESG) indicators on the financial performance of publicly listed entities in India. The findings highlight the importance of looking beyond the ESG score and assessing the most important ESG indicators that impact firms’ finan-cial performance. The main findings establish a relationship between Tobin’s Q (firm value) and individual ESG indicators. The study shows that ESG indicators have comparatively less impact on the internal accounting measures of Return on Total Assets (ROA) and Return on Net Worth (RONW). The results list the following key ESG indicators: energy intensity and consumption, equal and mini-mum wages to employees, turnover rate, as those having impact on Tobin’s Q across the various ML models explored. These research findings would assist corporations, investors, and policymakers in identifying key ESG indicators that impact financial performance of companies.
Authors - Siham Beloualid, Abderrahim BAJIT, Hicham Essamri, Youness Zahid, Nabil Srifi, Rachid Elbouayadi, Aya Essafini, Salaheddine Zezaan Abstract - The growing demand for intelligent healthcare systems emphasizes the need for advanced technologies that support continuous, non-intrusive patient monitoring in clinical environments. This paper presents a comprehensive solution that integrates both static and mobile IoT nodes, coupled with AI-driven perception and analytics, to enhance monitoring capabilities within hospital settings. The proposed system incorporates three core components: autonomous navigation, computer vision–based health assessment, and machine learning–enabled sensor analysis. Static nodes are deployed at each patient’s bedside to collect physiological and environmental data in real time using embedded sensors, while simultaneously capturing visual data through fixed cameras. Sensor data is analyzed using machine learning models that predict missing values in case of sensor malfunction or loss, ensuring reliable and uninterrupted monitoring. In addition to static setup, a mobile robot navigates autonomously through hospital wards, executing monitoring routines based on user-defined schedules or clinical priorities. During its rounds, the robot collects visual data of patients and surgical sites. Deep learning models, applied to extract subjective information, are employed to analyze patient posture, wound healing progress, and to detect essential items such as medicine containers and water bottles. These models also support autonomous navigation by enabling path recognition and action planning. The system integrates insights from both static and mobile nodes into structured, real-time reports that assist in the early detection of clinical issues and timely informed medical decisions. This solution provides a scalable and objective approach to continuous patient assessment. It improves monitoring quality services, reduces reliance on subjective manual evaluations, enhances staff efficiency, and fosters a more responsive, data-informed hospital environment.
Authors - Siham BELOUALID, Abderrahim BAJIT, Alae ELFOURCHI, Youness ZAHID, Emile SEMAKO, Hicham ESSAMRI Abstract - The increasing complexity of greenhouse agriculture requires integrated, scalable platforms capable of precise monitoring, real-time decision-making, and sustainable management. This work presents an intelligent greenhouse system that transitions from static IoT networks to a multi-agent architecture combining mobile sensing units, autonomous robots, and deep learning-powered computer vision. Environmental parameters such as temperature, humidity, and soil moisture are continuously collected through embedded sensors, while high-resolution visual data enables detailed plant health assessment. Machine learning models analyze these multimodal datasets to predict crop conditions, detect anomalies, and enable early intervention. A tri-modal vision system ensures complete coverage: static wide-angle cameras for global monitoring, mobile units with CNN-based analysis for close-range inspections, and robot-mounted cameras for real-time tasks. Autonomous robots, equipped with SLAM and path planning algorithms, perform precision operations such as targeted harvesting and disease removal. The fusion of robotics, AI, and sensor networks provides high-resolution monitoring, optimizes resource usage, and enhances crop management. Additionally, onboard edge computing enables fast, localized data processing, reducing latency and improving system autonomy.
Authors - Irna Hamzah, Ely Salwana, Hameedur Rahman Abstract - Virtual reality (VR) is swiftly evolving into a transformative tool in education, offering immersive environments that facilitate social-emotional learning (SEL) and enhance emotional regulation, particularly among adolescents. This study intends to explore the application of a VR design framework specifically customized for the Malaysian context, with the objective of strengthening social-emotional learning and establishing emotional regulation skills in adolescents. The study collects and analyses data through the VR application named mVR-REAL utilizing a VRSEL framework, which encompasses essential design principles such as user interface (UI), presentation layer (PL), and brain activity (BA) in the execution of VR-based social-emotional learning (SEL) interventions, highlighting the opportunities and constraints within Malaysian educational settings. The data analysis is performed utilizing SPSS. A total of 127 students from four secondary schools were enrolled in this study. Majority of the respondents were female (51.2%, n=65). All respondents answered usability instrument and the emotional questionnaire. Overall, the data suggests that the system has strong usability, with most users rating it as good. The mean score above 86% and the high proportion of good ratings reflect positively on the user experience. However, the presence of some users with scores below 80% highlights opportunities to further enhance usability for all users.
Authors - Rajneeshkaur Sachdeo Bedi, Pankaj Chandre, Pallavi Bhujbal, Bhagyashree Shendkar Abstract - Transforming healthcare with Natural Language Processing (NLP) is revolutionizing how medical data is analyzed, interpreted, and utilized. This paradigm shift is fueled by NLP's ability to process diverse healthcare data sources such as medical research articles, imaging reports, electronic health records, clinical notes, and patient feedback. Key NLP tasks like text summarization, information extraction, named entity recognition, sentiment analysis, and question answering are at the core of enabling advanced applications in healthcare. These include medical research acceleration, administrative task automation, clinical decision support systems (CDSS), and patient care enhancement. Despite its promise, challenges such as limited annotated datasets, biases in models, integration with legacy systems, and ensuring data privacy and security persist. The integration of NLP-driven solutions into healthcare systems has immense potential to enhance efficiency, improve patient outcomes, and optimize decision-making processes. Addressing these challenges through innovation and collaboration will ensure the sustainable growth of NLP in smart healthcare systems.
Authors - Khalid BOUALI, Abderrahim BAJIT, Hamza BENZZINE, Ibtissam EL AOUNI, Hicham ESSAMRI, Siham BELOUALID, Rachid EL BOUAYADI Abstract - The transition towards intelligent greenhouse agriculture is driven by the need to optimize resource efficiency and manage climate variability. Classical control methods, such as PID and open-loop control systems, have shown limi-tations in precision and in handling the nonlinearities. To address these chal-lenges, advanced automation and predictive control strategies like Model Predic-tive Control (MPC) have gained attention. However, achieving accurate forecasts of meteorological parameters remains critical for maintaining optimal growing conditions and reducing energy consumption. This paper proposes a hybrid deep learning model that integrates Long Short-Term Memory (LSTM) networks with Transformer architecture within an MPC framework to improve forecasting of key meteorological parameters, including temperature, humidity, and solar irra-diance. By combining memory and attention mechanisms, the model captures both short- and long-term dependencies in time series data. Experiments were conducted with varying input window sizes, and performance was evaluated us-ing mean absolute error (MAE), root mean squared error (RMSE), and symmetric mean absolute percentage error (SMAPE). Results show that the proposed hybrid model outperforms standard LSTM models in both accuracy and inference time, highlighting its potential to support intelligent greenhouse climate management and promote sustainable agriculture.
Authors - Khalid BOUALI, Abderrahim BAJIT, Hamza BENZZINE, Siham BELOUALID, Youness ZAHID, Mohamed Nabil SRIFI, Rachid EL BOUAYADI Abstract - With the rapid advancement of the Internet of Things (IoT), smart de-vices have become integral to various sectors, including healthcare, transporta-tion, smart homes, and industrial systems. In agriculture, these technologies have driven a shift toward automation and data-driven decision-making through the adoption of smart sensing systems. However, most existing systems remain con-strained by their reliance on centralized cloud infrastructure, resulting in latency, increased computational costs, and limited real-time capabilities. This work pro-poses an optimized microclimate sensor node architecture that incorporates edge computing abilities to enable real-time processing of sensor data and localized decision-making. These nodes are enhanced with Edge Artificial Intelligence to offer prediction analysis capacities at the edge. In this study, we focused on the prediction of missing sensor data using the intercorrelation between microclimate features, including temperature, humidity, CO₂, and light intensity. The study evaluates three supervised regression models, including Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Artificial Neural Network (ANN), based on R², RMSE, and MAE metrics. Experimental results demonstrate the superior performance of the RF model in learning the patterns between the mi-croclimate parameters, achieving R² scores of 98.88 % in predicting temperature, 95.60 % for humidity, 94.67 % for CO₂, and 97.40 % for light intensity. The results present the effectiveness of the model in predicting the missing parameters using the available ones. By enabling predictive inference at the edge, this ap-proach optimizes the monitoring systems, reduces the dependency on the cloud servers, and contributes to sustainable agriculture, generally offering a scalable solution for resilient, intelligent, and real-time monitoring in precision agricul-ture.
