Program

1.

Paper ID.32

Title: Identifying subgroups and behavioral profiles in Autistic children using Density-based clustering

Authors names:  Ahlam Belmaqrout, Btihal El Ghali and Najima Daoudi

Affiliation: School of Information Sciences, Rabat, Morocco

Abstract:

Autism spectrum disorder (ASD) exhibits a wide range of behavioral patterns, making it challenging to identify consistent subgroups across individuals. We used a dataset previously applied in a classification context to cluster diagnosed patients aged 4 to 16 years. The dataset considered for clustering contained 861 participants and included responses to Q-CHAT questions. Our proposed clustering approach ultimately identified 13–14 distinct subgroups that were stable, well-defined, and captured meaningful variations in response patterns. Some subgroups reflected subtle behavioral distinctions, while others represented broader patterns shared across multiple children, illustrating the spectrum of behavioral expression in ASD. These results provide a structured view of heterogeneity, showing that children can be grouped into reproducible and clinically meaningful profiles. The identification of these subgroups enhances understanding of behavioral diversity and offers practical implications for tailoring assessments and interventions to individual needs. Overall, this work highlights the value of detailed behavioral profiling, demonstrating that careful analysis of Q-CHAT responses can reveal robust and interpretable subgroups in ASD, supporting personalized approaches to therapy.

2.

Paper ID. 34

Title:   Predictive Medicine in Cardiology using AI and OR Approaches: A Systematic Review

Authors names: Khalid El Yassini*, Chaimaa Oufkiri*, Hafida Baazizi, Maimouna Ball* and Kenza Oufaska**

Affiliations:  *Moulay Ismail University, Faculty of Sciences Meknes, Morocco

                       **International University of Rabat, Morocco

Abstract:

The convergence of artificial intelligence (AI) and operations research (OR) in predictive medicine in cardiology is gaining increasing attention for its capacity to process complex data and its potential to improve the prediction and management of cardiovascular diseases. This study presents a systematic literature review (SLR) conducted using the Scopus database to examine how AI and OR contribute to advancements in this field. The findings indicate that Machine Learning (ML) and Deep Learning (DL) dominate the AI approaches utilized, applied to various types of medical data, including electronic health records (EHRs), imaging, genetic information, and physiological signals. AI-based models exhibit considerable advancements in risk stratification, disease progression modeling, and patient-specific treatment recommendations. The review highlights that the most commonly used evaluation metrics include accuracy, sensitivity, specificity, and area under the curve (AUC) scores, reflecting a strong focus on predictive performance. Additionally, studies emphasize the necessity for explainability and generalizability in AI-driven clinical

decision support systems. However, while AI-driven methods are well represented, OR is rarely mentioned in the analyzed studies. This review aims to guide future research toward a more comprehensive integration of these technologies in predictive medicine.

3.

Paper ID. 52

Title: Towards Resilient Smart Ecosystems: A Kubernetes-based AIOps Environment for Microservices Failure Prediction

Authors names: Ilyass Tarhri, Driss Allaki and Hamza Kamal Idrissi

Affiliation: National Institute of Posts and Telecommunications, Rabat, Morocco

Abstract

With the growing dependence of digital infrastructure on cloud-native microservices and Kubernetes, maintaining reliability has emerged as a major challenge. These environments create intricate, non-linear failure patterns that standard monitoring techniques frequently overlook. Although AIOps presents a possible solution, the creation of efficient predictive models is presently constrained by a lack of high quality training data. This paper outlines a taxonomy of Kubernetes failure types and suggests a reference framework for an experimental environment. We combine the intricate Train Ticket microservices benchmark with the Chaos Mesh fault injection framework to make a controlled environment that can make multi-modal, ground-truth datasets. This proposed architecture addresses current data limitations, establishing a fundamental framework for the forthcoming training and evaluation of next-generation reliability models in distributed systems.

4.

Paper ID. 53

Title : Cirrhosis Risk Assessment using Machine Learning for Proactive Healthcare Management

Authors names: Shaista Farhat*, Sumaiya Shaikh*, M.A. Jabbar* and Abdelkrim Haqiq**

Affiliations: *Vardhaman college of Engineering Shamshabad, Hyderabad, India

                      **Hassan 1^st University, Faculty of Sciences and Techniques, Settat, Morocco

Abstract:

In India, cirrhosis is a major health issue that contributes to high rates of morbidity and mortality. The prevalence of cirrhosis in India is caused by a number of variables, including alcohol intake, fatty liver disease, metabolic problems, and viral hepatitis infections (particularly hepatitis B and C). Severe problems such as portal hypertension, ascites (a build-up of fluid in the abdomen), hepatic encephalopathy (a disease affecting brain function), variceal hemorrhage, and an elevated risk of liver cancer can arise as cirrhosis advances. Early on, symptoms may not be noticeable which cause delayed diagnosis. Many times, complications have already arisen when cirrhosis is detected at an advanced stage, which limits available treatments and worsens prognosis. Nowadays Machine Learning techniques to the healthcare industry has showed promise in recent years for transforming proactive disease management and risk assessment. A wide range of datasets, including clinical records, test findings, and demographic data from patients with and without cirrhosis, are used in this study. To extract significant patterns and create predictive models, feature engineering approaches and sophisticated machine learning models—such as ensemble methods and deep learning architectures—are used. In proposed model XG Boost and Cat Boost Classifier were used which worked outstanding with an accuracy of 96.18%. The system's functions include predicting the likelihood of cirrhosis, classifying patients according to risk, and recommending specific therapies for those who are considered high-risk.

5.

Paper ID. 54

Title: AI Agent-based Social Trend Intelligence turns noise into e-commerce ready fashion signals to drive personalized consumer behavior

Authors names:  Sripriya Thinagar*, Sathiya Pandi*, Jayashree Ganeshkumar** and Siddhika Sriram*

Affiliations: *ManoloAI, Texas, USA

                      ** Avinashilingam Institute for Higher Education for Women, Coimbatore

Abstract:

Social trend intelligence leveraging a hashtag agent-based analysis plays a critical role in identifying emerging fashion trends by leveraging structured social media data and complex network analysis techniques. Consumer behavior has pushed retail e-commerce to accelerate the production and sale of high fashion trending designs. To meet this changing consumer dynamic, manual exploration of hashtags is costly, time consuming and inconsistent, making a systematic computational approach more reliable and scalable. The Hashtag Agent ingests a predefined set of seed hashtags stored in a relational database and scrapes social media posts from multiple sources including a NoSQL data source. Related hashtags are extracted and analyzed using co-occurrence score calculations to measure how frequently two hashtags appear together within the same post. A weighted hashtag network model is then constructed based on co-occurrence relationships to represent the strength of connections between hashtags. Louvain community detection is applied to this network to identify meaningful clusters representing trend groups within the fashion domain and support subsequent data analytics. These clusters help distinguish strong thematic communities and highlight emerging patterns across categories and styles. This approach ensures scalable, data-driven trend expansion while reducing duplicate scraping and improving analytical accuracy. The findings demonstrate that combining co-occurrence analysis with Louvain clustering provides an effective and interpretable framework for continuous fashion trend discovery and strategic decision support.