10th International Congress on Information and Communication Technology in concurrent with ICT Excellence Awards (ICICT 2025) will be held at London, United Kingdom | February 18 - 21 2025.
Sign up or log in to bookmark your favorites and sync them to your phone or calendar.
Authors - Indika Udagedara, Brian Helenbrook, Aaron Luttman Abstract - This paper presents a reduced order modeling (ROM) approach for radiation source identification and localization using data from a limited number of sensors. The proposed ROM method comprises two primary steps: offline and online. In the offline phase, a spatial-energetic basis representing the radiation field for various source compositions and positions is constructed. This is achieved using a stochastic approach based on principal component analysis and maximum likelihood estimation. The online step then leverages these basis functions for determining the complete radiation field from limited data collected from only a few detectors. The parameters are estimated using Bayes rule with a Gaussian prior. The effectiveness of the ROM approach is demonstrated on a simplified model problem using noisy data from a limited number of sensors. The impact of noise on the model’s performance is analyzed, providing insights into its robustness. Furthermore, the approach was extended to real-world radiation detection scenarios, demonstrating that these techniques can be used to localize and identify the energy spectra of mixed radiation sources, composed of several individual sources, from noisy sensor data collected at limited locations.
Authors - Shima Pilehvari, Wei Peng, Yasser Morgan, Mohammad Ali Sahraian, Sharareh Eskandarieh Abstract - Overfitting is a common problem during model training, particularly for binary medical datasets with class imbalance. This research specifically addresses this issue in predicting Multiple Sclerosis (MS) progression, with the primary goal of improving model accuracy and reliability. By investigating various data resampling techniques, ensemble methods, feature extraction, and model regularization, the study thoroughly evaluates the effectiveness of these strategies in enhancing stability and performance for highly imbalanced datasets. Compared to prior studies, this research advances existing approaches by integrating Kernel Principal Component Analysis (KPCA), moderate under-sampling, Synthetic Minority Oversampling Technique (SMOTE), and post-processing techniques, including Youden’s J Statistic and manual threshold adjustments. This comprehensive strategy significantly reduced overfitting while improving the generalization of models, particularly the Multilayer Perceptron (MLP), which achieved an Area Under the Curve (AUC) of 0.98—outperforming previous models in similar applications. These findings establish important best practices for developing robust prognostic models for MS progression and underscore the importance of tailored solutions in complex medical prediction tasks.
Authors - Ain Nadhira Mohd Taib, Fauziah Zainuddin, M. Rahmah Abstract - This paper presents AdaptiCare4U, an interactive mental health assessment in high school settings. By integrating adaptive technique with an establish mental health assessment instrument in a user-friendly format, Adap-tiCare4U improves the experience in answering mental health assessment. Through expert review validation technique, AdaptiCare4U demonstrates high effectiveness in accessibility, ease of use, and practical value with mean scores of 5, 4.2, and 4.4 respectively. Additionally, students’ perception further supports the tool’s usability, with positive feedback highlighting its engaging interface, use of multimedia elements, and stress-reducing design. A favorable usability rating from both students and experts makes AdaptiCare4U a promising tool for aiding counselors in conducting efficient mental health assessments.
Authors - Aayush Kulkarni, Mangesh bedekar, Shamla Mantri Abstract - This paper proposes a novel serverless computing model that addresses critical challenges in current architectures, namely cold start latency, resource inefficiency, and scalability limitations. The research integrates advanced caching mechanisms, intelligent load balancing, and quantum computing techniques to enhance serverless platform performance. Advanced distributed caching with coherence protocols is implemented to mitigate cold start issues. An AI-driven load balancer dynamically allocates resources based on real-time metrics, optimizing resource utilization. The integration of quantum computing algorithms aims to accelerate specific serverless workloads. Simulations and comparative tests demonstrate significant improvements in latency reduction, cost efficiency, scalability, and throughput compared to traditional serverless models. While quantum integration remains largely theoretical, early results suggest potential for substantial performance gains in tasks like function lookups and complex data processing. This research contributes to the evolving landscape of cloud computing, offering insights into optimizing serverless architectures for future applications in edge computing, AI, and data-intensive fields. The proposed model sets a foundation for more efficient, responsive, and scalable cloud solutions.
Authors - Nouha Arfaoui, Mohmed Boubakir, Jassem Torkani, Joel Indiana Abstract - The increasing reliance on surveillance systems and the vast amounts of video data have created a growing need for automated systems to detect violent and aggressive behaviors in real-time. Manual video analysis is not only labor-intensive but also prone to errors, particularly in large-scale monitoring situations. Machine learning and deep learning have gained significant attention for their ability to enhance the detection accuracy and efficiency of violence in images and videos. Violence is a critical societal issue, occurring in public spaces, workplaces, and social environments, and is a leading cause of injury and death. While video surveillance is a key tool for monitoring such behaviors, manual monitoring remains inefficient and subject to human fatigue. Early ML methods relied on manual feature extraction, which limited their flexibility in dynamic scenarios. Ensemble techniques, including AdaBoost and Gradient Boosting, provided improvements but still required extensive feature selection. The introduction of deep learning, particularly Convolutional Neural Networks (CNNs), has enabled automatic feature learning, making them more effective in violence detection tasks. This study focuses on detecting violence and aggression in workplace settings by addressing key aspects such as violent actions, and aggressive objects, utilizing various deep learning algorithms to identify the most efficient model for each task.
Authors - Kalupahanage A. G. A, Bulathsinhala D.N, Herath H.M.S.D, Herath H.T.M.T, Shashika Lokuliyana, Deemantha Siriwardana Abstract - The explosive growth of the Internet of Things (IoT) has had a substantial impact on daily life and businesses, allowing for real-time monitoring and decision-making. However, increased connectivity also brings higher security risks, such as botnet attacks and the need for stronger user authentication. This research explores how machine learning can enhance Internet of Things security by identifying abnormal activity, utilizing behavioral biometrics to secure cloud-based dashboards, and detecting botnet threats early. Researchers tested numerous machine learning methods, including K-Nearest Neighbors (KNN), Decision Trees, and Logistic Regression, on publicly available datasets. The Decision Tree model earned an impressive accuracy rate of 73% for anomaly identification, proving its supremacy in dealing with complex security risks. Research findings show the effectiveness of these strategies in enhancing the security and reliability of IoT devices. This study provides significant insights into the use of machine learning to protect Internet of Things devices while also addressing crucial concerns such as power consumption and privacy.
