CENTRAL ASIAN JOURNAL OF MATHEMATICAL THEORY AND COMPUTER SCIENCES

Analysis of Probability and Statistics Topics in Mathematics Textbooks (Singapore, Japan, USA, Russia)

Yusubjanova Musharraf — 2025-08-02

Teaching the subject of mathematics has always been considered relevant in every era. Mastering this subject well not only affects students' knowledge but also has a significant impact on indicators that determine their future. Especially, the knowledge related to the sections of Probability Theory and Statistics in mathematics is being used in many fields. For example, constructing models of research conducted in all disciplines, evaluating them, showing their validity, and making forecasts all require knowledge from this subject. In addition, Probability Theory and Statistics help develop logical reasoning, gain profits in business through risk-taking, and choose the option with the highest probability among different life choices. For instance, the world's major investors rely on probability and statistical data when investing in a field, country, or entrepreneur that promises good returns. This article analyzes the teaching of Probability Theory and Mathematical Statistics sections in general education school textbooks through the experiences of foreign countries (Singapore, Japan, USA, Russia). That is, according to the curriculum, the topics of Probability Theory and Mathematical Statistics are given in mathematics textbooks for grades 7–11, and the composition, sequencing, and interconnection with previous and subsequent topics are explained. The diversity and real-life relevance of the problems presented in textbooks are also discussed.

Between Human Trust and Algorithmic Prediction: Designing Intelligent Systems Using XAI and Modern Big Data Platforms

Ali Hussein Khalaf Al-Sammerraie — 2025-08-09

Big data and AI have changed how organizations make decisions, and businesses rely more and more on automated decision-making to achieve better performance, manage risks, and develop and implement business strategies. But complexity and opacity of AI models can often erode user trust, particularly in high trust industries like finance, health and customer analytics. To solve this problem, This research engineers an intelligent analytical system that bridges algorithmic accuracy and human interpretability by integrating XAI with big data platforms. The framework addresses scalability challenges in industrial environments through a layered architecture optimized for real-time deployment." that bridges. We construct a layered architecture, with data ingestion and pre-processing using Spark and Pandas; model training; explainability; and a user-friendly interface using Streamlit and FastAPI. We validated the proposed approach through a real case concerning customer churn prediction in order to show that it provides competitive predictive accuracy, as well as transparent decisions. The experimental results indicate that SHAP enhances the user’s understanding and trust in AI-based decisions. This work provides a scalable, interpretable, and practical approach that is UAD-based for the deployment of intelligent decision support systems in enterprise settings, adding to the literature on trustful AI.

AI Voice Assistant and Caption Generation Using Convolution Neural Network and Bi LSTM

G. Rajasekaran — 2025-08-07

This research examines the physiological reactions to stress in light of the heightened stress levels observed among workers and students during the COVID-19 pandemic, particularly in distant work and virtual learning contexts. The study investigates several factors that affect stress levels in distant workers and students since it is important to understand these responses. By using ideas from psychology, physiology, and technology, the research finds the main causes of increased stress in different groups of people. The suggested approach has important effects on occupational health because it gives remote workers access to tools and information that can help them make their work environments healthier. In the same way, the system improves student well-being in virtual learning environments by giving them important support during the difficulties of remote learning. Additionally, this type of technology is useful in other areas, such as telemedicine, and it is helping to create technology-based solutions for managing stress and improving health in general. We want to help people deal with the stress of working from home and learning online by using new technologies and greater understanding of psychology. In the end, we want to help people become more resilient and healthy in the face of new obstacles.

Strategic Implementation of IT Governance for Optimising Information Systems Performance

Zahraa Sameer Ibrahim — 2025-08-09

This article presents a new reinforcement learning approach to optimise the use of IT governance resources to enhance information system performance. Despite enormous IT infrastructure investments, organisations might fail to recoup the investment in terms of quantifiable performance enhancements due to ineffective governance practices. We developed a Q-learning system that learns dynamic optimal policies of resource allocation over the most significant five IT governance dimensions: Strategic Alignment, Value Delivery, Resource Management, Risk Management, and Performance Measurement. Using ESG data sets of 12 countries with various levels of digital maturity, our system learned governance dimensions yielding maximum return on investment at every level of maturity. Results indicate that the emerging economies achieve optimal performance improvements of as much as 99% through emphasis on Strategic Alignment and Resource Management, while developed economies achieve this through more balanced distributions with greater emphasis on Risk Management. The Pareto frontier analysis confirms that our Q-learning method reaches optimal levels of resource utilisation efficiency. Cluster analysis determines sharp-cut modes of governance by levels of economic development. This research offers an evidence-based adaptive approach to IT governance implementation that takes into consideration organization maturity and context to enable decision-makers to deploy scarce resources in order to realise peak information systems performance.

Designing an Optimized Algorithm for Cyberattack Detection in IoT Systems

Noor Jihad Kadhim — 2025-08-07

The Internet of Things (IoT) connects billions of devices, which makes them helpful in many areas. But the fast growth of linked nodes is a big security risk, especially spoofing attacks, which modify the IDs of Pong Humans devices and make the network less trustworthy. This research provides a novel and better way to uncover spoofing attacks in IoT environments using the Random Forest (RF) model. We use fake datasets that act like IoT traffic and threats to train and test the system. We assess the suggested model using performance metrics including accuracy, precision, recall, and F1-score. We compare our work to various classifiers, like K-Nearest Neighbor (KNN) and Support Vector Machine (SVM). The results show that the optimized RF model is better than the others because it has a 96.8% accuracy rate and can find more spoofing attempts. The study introduces a lightweight and scalable method that performs well in IoT settings with less resources.

Study Compression with Different New Prior distribution in Tobit Quantile Regression

Abbas Shaker Mohsen Al-Jashami — 2025-08-11

Bayesian estimation requires sampling from the posterior distributions. Where, the prior distributions are play a vital role in obtaining the simplifying the derivation of full conditional distributions, making Gibbs sampling algorithms more efficient. using the Laplace prior distribution (also known as the Double Exponential prior) is indeed a great choice in Bayesian Tobit quantile regression for both variable selection and parameter estimation simultaneously. The Laplace prior has become popular in regression models because of its ability to induce sparsity in the estimated coefficients, which is particularly beneficial for variable selection. However, directly using the Laplace prior distribution is a very complex task when building the hierarchical model. To overcome this issue, a set of transformations of the Laplace prior distribution has been used, which provide us with hierarchical models with more efficiency. In this paper, we will compare the transformations of the Laplace prior distribution that provide us with efficient estimators capable of generalization.

Smart Book Hub: Empowering Literary Discovery Experience, Tailored Book Recommendations Using Mern and Lamp Stack

S Ramesh Kumar — 2025-08-20

Every new technology that comes out has some new features that meet the needs of users. A lot of innovations come and go from the market, but some have a big effect, make big improvements, and stay on top. The proposed online marketplace for used books and textbooks, which was made in PHP, includes user authentication, a strong system for uploading books with categories like business, sports, management, tourism, informatics, accounting, applied science, and more. It also includes important information like the ISBN, condition, price, and cover image. Users can browse the platform to see what books are available, add them to their shopping cart, and then go through a safe checkout process. The app has a user dashboard where you can manage your books, see your order history, and change your profile. To protect against any weaknesses, security procedures like input validation, using HTTPS, and storing passwords securely are put in place. You can add a payment gateway to the platform if you want to. In general, the system puts a responsive design first so that it works on all devices, and it comes with full documentation for future maintenance. We use the MERN (MongoDB, Express.js, React.js, Node.js) stack for temporary server functionality and the LAMP (Linux, Apache, MySQL, PHP) stack for home server operations. This gives us the best of both worlds.

Convolutional Neural Network-Based Classification Models for the Detection of Diabetic Retinopathy in Retinal Fundus

Zainab Fahad Alnaseri — 2025-08-19

Diabetes mellitus is a widespread concern worldwide and the most serious microvascular diseases that usually occur as a result are those of the eye, which include the retinopathy and macular edema. Over the past decade, DR has emerged as a significant player in causing vision disability and blindness. Provided that diabetes-related eye complications are promptly diagnosed and handled, the consequences of them may be significantly improved and the level of sugar in the blood can be kept on an adequate level. Nevertheless, the DR symptoms are not consistent and may be complicated; thus, doctors may spend a lot of time to diagnose them.One of the approaches to detecting and classifying DR on fundus retina photographs that is taken into account in the paper is the one which relies on CNNs and deep learning. All the experimental data used in the present study was taken at the Department of Ophthalmology at Xiangya No. 2 Hospital at Changsha in China. The sample of cases is not considerable and the information included in this dataset is imbalanced. That is why a system was made that can be used to rectify the variety and excellence of the information utilized in the training by normalizing and creating information.Then, many CNNs such as "ResNet"-101, "ResNet"-50, and "VGGNet"-16 were employed to ascertain the phases of DR. "ResNet"-101 outperformed the other models by getting 98.88% accuracy and losing 0.3499 during training and 0.9882 during testing. The model was checked on datasets such as HRF, STARE, DIARETDB0, and XHO, which contain 1,787 examples and resulted in an average accuracy of 97%, making it higher than existing methods on the same subject. As a result, using this proposed model enhances DR detection accuracy more than "ResNet"-50 and "VGGNet"-16, making it promising for DR screening in health services.

Walrus Optimizer Based Novel Energy-Efficient Clustering for Wireless Sensor Network

Hind E. Mohammed — 2025-08-24

Wireless sensor networks (WSNs) have important role in modern Internet of Things (IoT) systems also make effective set of data as well as transmitting able. Although, energy sources restrictions in sensor nodes refers to basic concern, particularly in big-scale networks. Techniques based on clustering were presented as efficient concern for raising effectiveness of energy, however optimum clusters’ shape is yet the essential study concern. In this paper, the new mechanism of clustering given the Walrus Optimizer algorithm is defined. Such algorithm, inspired by group walruses manner, optimizes formation and head selection cluster with decreasing energy use and balanced load share objective between nodes. Outcomes of simulation illustrate that presented technique given the Walrus Optimizer performs better than traditional algorithms like Genetic Algorithms (GA) also Particle Swarm Optimization (PSO) in case of network life and energy effectiveness. Specifically, the technique develops network lifetime by 13.45% and decreases use of energy by 10-15% in comparison with base techniques. Present study results illustrate Walrus Optimizer algorithm ability to solve main WSNs issues and paves the way for later study in applying nature-inspired mechanisms’ domain in WSNs.

Bayesian Reciprocal LASSO Composite Quantile Regression for Robust Clinical Risk Modeling

Ali Abdulmohsin Abdulraeem Al-rubaye — 2025-08-25

Clinical data often contain outliers and irrelevant predictors that can distort inference and reduce the reliability of traditional regression methods. To address this issue, we propose a robust Bayesian variable selection framework by integrating composite quantile regression with a reciprocal LASSO prior. The method accommodates heavy-tailed errors and performs simultaneous coefficient estimation and sparsity enforcement.We evaluate the proposed model through extensive simulation studies under contamination scenarios and compare it with classical and Bayesian LASSO-based quantile regression methods. The model is further applied to systolic blood pressure data from the NHANES 2017–2018 survey to identify key lifestyle and health-related predictors. Results show that the proposed method outperforms competing approaches in terms of predictive accuracy, robustness to outliers, and variable selection stability.

Some Results on Rough k-Space

Mudheher Abdul Jabbar Hasan Albayati — 2025-08-29

This paper introduces the concept of rough k-space within the framework of rough set theory. The primary aim of this work is to define rough k-space and explore its properties, including rough continuity, homeomorphisms, and topological characteristics. Specifically, it is shown that the restriction of a rough continuous function to any rough compact subset of a rough space remains rough continuous. Additionally, the cross product of a rough k-space with a rough compact T₂-space results in a rough k-space. The study also highlights key hereditary and topological properties of rough k-spaces. The novelty of this research lies in its extension of rough set theory to include the concept of rough k-spaces, which integrates topological and rough set properties, and introduces a new approach to understanding the interaction between rough sets and continuous functions. Furthermore, the paper provides detailed results on the continuity and homeomorphism properties of rough k-spaces, offering a fresh perspective on their application in mathematical and computational contexts. The implications of these findings are significant for further research in rough topology, particularly in the development of robust mathematical models for rough set theory and its applications in areas such as decision-making, knowledge discovery, and artificial intelligence.

Comparison of Cox Proportional Hazards Regression and Self-Supervised Learning Algorithm in Estimating Lung Cancer Risk

Enas Abid Alhafidh Mohamed — 2025-08-29

This research compares three models for analysing survival data for lung cancer patients: the Cox proportional hazard model, the supervised self-learning algorithm, and a hybrid model that combines the best parts of the two models. Using MatLab, the comparison was made using multiple performance assessment criteria, such as the mean absolute error (MAE), the mean square error (MSE), the accuracy index (C-index), and Akaike's criterion (AIC). The hybrid model was more accurate than the baseline models, with an accuracy of 0.94 and reduced comparison criteria. The Cox model, on the other hand, only had an accuracy of 0.82. The risk data from the sample also indicated that advanced disease stage, smoking, age, and being male were the factors that most elevated the risk of lung cancer. On the other hand, immunotherapy and radiation lowered the risk of lung cancer. So, the hybrid model is a good way to figure out how likely someone is to die.

Review of Ethical Implications of AI: Balancing Innovation and Privacy in Digital Societies

Rusul J. Alsaedi — 2025-08-31

Artificial intelligence (AI) has a big impact on how digital societies are formed in many areas, especially in research, government, and healthcare. AI also has to do with keeping personal information safe. This study looks at the moral issues that come up with AI. It stresses protecting basic rights and encouraging new ideas. This study looks at both sides of AI, including its ability to change things and the worries that come with it, as well as algorithmic decision-making and community surveillance. This study looks at modern ethical frameworks and governance models, stresses global efforts, and adds trust to AI systems. It stresses how important it is to protect users' privacy and calls for ethical AI. This study says that we should promote and improve technology in a way that is fair and focuses on people.

Artificial Intelligence in Higher Education: Enhancing, Teaching and Learning through Adaptive Technologies

Rusul J. Alsaedi — 2025-09-01

Artificial Intelligence (AI) is moving the higher education from state to another state by using private, data-driven, and measured learning involvements. This paper discusses the combination of AI essentially usual technologies into the university level of education. It studies the function of intelligent teaching systems, virtual assistants of teaching, and adaptive programs of learning in developing the engagement of student, academic attitude, and instructive adequacy. Although the advances of AI in the higher education system are important, such as enlarged accessibility and tailored education ways, these benefits also show the ethical sides such as information secrecy, algorithmic alignment, and the changeful turn of human teachers. Depending on the recent studies, this article discusses that reliable selection of AI needs institutional aid, faculty practice, and powerful schema structures. Finally, AI has the ability to aid more universal and active educational environments when performed with impressionability, equity, and educational notion.

Modern Trends in Using Educational Technology to Develop Mathematics Learning: A Review of Arabic Studies from 2019 to 2024

Ameer Hadi Obada — 2025-09-03

Mathematics is a fundamental discipline that fosters analytical thinking and problem-solving skills, while also supporting technological innovation and national development. However, many learners perceive it as abstract and disconnected from real life, which negatively affects engagement and achievement. In recent years, educational technology has emerged as a key strategy to bridge this gap by enabling interactive, meaningful, and student-centered learning experiences. This study provides a descriptive analytical review of Arabic research on technology-enhanced mathematics education published between 2019 and 2024. Data were collected from peer-reviewed journals across Jordan, Iraq, Libya, and Egypt, focusing on interventions that reported cognitive and affective learning outcomes. The analysis examined types of technology, their pedagogical roles, and resulting student outcomes. Findings revealed consistent evidence that digital tools, including computer-assisted instruction, e-learning platforms, and game-based applications, significantly improve conceptual understanding, problem-solving, motivation, and engagement. Moreover, the effectiveness of these interventions depends strongly on teacher proficiency, digital infrastructure, and the alignment of technology with curricular standards. While technology enhances both achievement and attitudes towards mathematics, its impact is maximized when integrated within supportive pedagogical strategies. This review highlights the potential of technology to transform mathematics education in the Arab context and provides practical implications for educators, researchers, and policymakers seeking to optimize its integration.

Applications of Sustainable Transportation and AI Models for Regional Economic Growth Prediction

Khalid zeghaiton chaloob — 2025-09-03

The interplay role between sustainable transportation infrastructure and regional economy is indispensable. It accentuates the transformative action of artificial intelligence (AI) in predictive economic modeling. Analyzing the economic impacts on the transport systems (highway, railway, maritime, and airways) highlights how transport modes deepen the connectivity, reduce costs, and stimulate growth in various industries. Cost-Benefit Analysis (CBA) and Computable General Equilibrium (CGE) are frameworks that frequently implemented to evaluating empirical models. Advanced AI models explored covering transformer models and federated learning to forecast economic trends with higher accuracy. Four case studies reviewed from regions including Brazil, India, South Africa, and Japan. The findings emphasize essential integrating of AI with traditional statistical models addressing data complexity and improving policy making. Our contribution is to present the adopted sustainable development and technological innovation in economic planning.

Smart Hydroponic Growth Optimisation System with Real–Time Monitoring and Control with Convolutional Neural Network Algorithm Using Machine Learning

G. Rajasekaran — 2025-09-07

This project aims to develop a smart hydroponic plant monitoring system utilising a range of sensors and actuators to enhance the environment for plant growth. The system automatically changes environmental conditions based on real-time data collected from sensors. This makes hydroponic farming more productive and efficient. The Light Dependent Resistor (LDR) detects light intensity, the DHT11 checks the temperature, and the PH sensor checks the quality of the water. Light, temperature, water pH, and nutrition levels are all controlled by actuators such as bulbs, fans, water pumps, and motors. The main goal of this system is to create an automated and efficient method for monitoring hydroponic plants, thereby enabling them to develop more effectively and reducing the need for human intervention. The study also uses deep learning methods to find diseases in spinach leaves, which adds another layer of plant health monitoring to the system. This combination of technology and software makes it easier to run hydroponic farms and also increases their overall efficiency and output. The smart hydroponic plant monitoring system in this project is a paradigm for sustainable farming. It shows how technology can be used to make farming more effective and environmentally benign. This technology aims to enhance hydroponic farming methods by automating the monitoring and adjustment of key growth factors. This will lead to better crop yields and more efficient use of resources.

Multimodal Deep Learning for Enhanced Stock Market Trend Prediction

Basma Hussein Hameed — 2025-09-10

Stock market prediction is an important tool in investment decision, optimization of portfolio, and management of risks. The historical price and economic indicator paradigm of traditional forecasting approaches would fail to include the sweeping and quick movements of behaviors in a volatile market, thus lacking predictive capability . The social media, including Twitter and Reddit, have proved to be a rich source of investor sentiment, with real-time accounts of what the public expects and how the market is being shaped by psychological drivers . Yet, the majority of the available literature relies on a single dimensional research design, i.e., investigating either sentiments in a text or economic measures, therefore, omitting the complementary relationship between the two dimensions. The purpose of the current study is to build a strong multimodal deep learning model that would incorporate the investor sentiment in the social media with other aspects of economic indicators. The goal is to provide more trustworthy, consistent, and interpretable predictions concerning the stock market trend and allow investors and financial institutions to make more informed decisions. The hypothesized framework comprises two major branches: (1) a Transformer-based model of sentiment analysis (FinBERT and RoBERTa) to make context-informed embeddings out of social media posts and (2) an LSTM-based branch to infer sequential implications of economic factors such as interest rates, trading volumes, and inflation. A late-fusion approach combines and trained to discover cross modal connections across both branches before being classified as either an upward, downward or neutral tendency. The Twitter, Reddit and StockTwits Twitter data were used with the economic data of Yahoo Finance and FRED over 2023-2025. The performance was measured in Accuracy, Precision, Recall, F1-score, and ROC-AUC.The multimodal model fared much better than its unimodal counterparts with an Accuracy of 91.2%, F1-score of 90.7 and ROC-AUC of 0.94. To ensure that such improvements are not random, paired t-tests and ANOVA was used to prove that any such improvements were verified as significant statistically (p < 0.05). The sentiment data had a larger effect on the short-term forecasts, whereas the economic indicators aided in the long-term stability, which proves the complementarity of the behavioral and the fundamental data.

An Improved Numerical Analysis for Solving Nonlinear Equations Using a Hybrid Algorithm Between Newton’s Method and Metaheuristic Approaches

Hussein Dhahir Habeeb Alaid — 2025-09-23

The problem of finding all roots of a system of nonlinear equations is a common difficulty in scientific computing and various engineering disciplines. Classical numerical methods like Newton’s method and Levenberg-Marquardt usually have drawbacks, sensitive to initial guesses, local convergence, etc., and are not able to find multiple solutions. This paper explores the potential of metaheuristic algorithms, and more specifically Genetic Algorithms (GAs), for handling these challenges. The study shows that GAs represents a powerful and versatile solution to solve linear and nonlinear system, with the ability to explore the global solution space, avoiding local minima, and the capability of converging to the correct solution from poor initial guesses. A complementary strength of GA is that it can retrieve multiple solution sets, something deterministic methods cannot do as well (especially in systems where there are multiple physically meaningful equilibria). The behavior of the GA is examined on a series of benchmark problems and compared to traditional methods, i.e., Gaussian Elimination, and Newton's scheme. . It is established that GA will provide the exact solutions for the determination of the global optimum and is superior compared to other optimisation algorithms for the solution of complex and multimodal spaces of solutions. Also, the analysis reveals the possibility of the hybrid form of the metaheuristics with the local refinement for the target of minimising convergence and augmenting precision. The analysis highlights the need for the evolutionary analysis for numerical analysis and suggests the integration of the usage of the methodology of the metaheuristic within the process of the solutions for the complex mathematical problems. The results are especially significant for optimisation, control and computational physics problems where reliability and diversity can be crucial among solutions.

Micro Irrigation by Integrating AI to Predict Crop Water Needs and Automate Valves and Boost Yield

R Sivakani — 2025-09-23

This study suggests combining artificial intelligence (AI) with micro irrigation technologies to improve how farmers manage water for their crops. Micro irrigation systems give water to crops very accurately, but they have trouble adjusting to changes in the weather and the water needs of the crops. This project wants to use AI algorithms to figure out how much water crops will require and to automatically open and close valves based on weather forecasts. The predictive modelling part is about making AI algorithms that can use information like the type of crop, the amount of moisture in the soil, and the weather to guess how much water the crops will need. The model will be trained using machine learning approaches including regression analysis and neural networks. The model's accuracy will keep getting better as it gets more historical and real-time data. To make the micro irrigation system automatic, smart valves, sensors, and actuators will be added. The AI algorithm and real-time weather data will tell these valves when to open and close. The goal of this dynamic adjustment is to make water delivery as efficient as possible while wasting as little as possible. Some of the main benefits are that it can respond to weather events before they happen, it can grow, and it works with current farming methods. User-friendly interfaces will make it possible to monitor and control things from afar. In conclusion, combining AI with micro irrigation systems is a promising way to make farming more environmentally friendly, use water more efficiently, and grow more crops.

Aligning Marketing and Supply Chain Functions: A Longitudinal Study on the Impact of Cross-Functional Integration Mechanisms on Market Responsiveness and Firm Profitability

Golan Muwafaq Abdullah Abdullah — 2025-09-23

A fundamental and persistent challenge in organizational management is the inherent misalignment between marketing and supply chain functions. Historically, marketing focuses on demand generation through product variety, promotion, and availability, while supply chain management prioritizes efficiency in demand fulfillment through cost minimization, standardization, and stable operations. This natural divergence in goals and incentives often creates conflict, operational inefficiencies, and suboptimal overall performance, ultimately eroding competitive advantage. Although the positive correlation between integration and performance is established, extant literature is predominantly cross-sectional, capturing a static snapshot of this relationship. A significant gap exists in understanding the temporal evolution of this dynamic—specifically, how different integration mechanisms develop and compound their effects on performance outcomes over time, and through what mediating processes. This study aims to address this gap by employing a longitudinal research design to examine the impact of specific cross-functional integration mechanisms—categorized as structural (e.g., joint planning, profit centers), technological (e.g., integrated information systems), and social (e.g., trust, collaborative culture)—on both market responsiveness and firm profitability. Data were collected from senior managers over a three-year period.The results indicate that both formal and informal integration mechanisms have a significant positive impact on a firm's market responsiveness. Furthermore, the effect of integration on profitability is found to be partially and, in some cases, fully mediated through the enhancement of market responsiveness. Longitudinal analysis reveals that these positive effects are not immediate but strengthen and become more significant over time, underscoring the evolutionary nature of capability building. This research makes two primary contributions. Theoretically, it provides longitudinal empirical evidence and tests a mediation model, offering a more nuanced understanding of how and over what time period integration influences profitability. Practically, it provides managers with an evidence-based roadmap, identifying the most effective integration mechanisms to prioritize for improving long-term performance.

Mathematics Applications to Detect Fetal Syndromes

Faten Hameed Sabty — 2025-09-24

This research addresses the detection of fetal syndromes as a high-dimensional, non-linear binary classification problem, we mathematically formulate and empirically evaluate three classes of models: probabilistic classifiers based on Bayesian inference with multivariate Gaussian assumptions, geometric classifiers such as Support Vector Machines with non-linear kernels, and deep learning models based on multi-layer neural networks, the study's central hypothesis posits that the complex, synergistic interplay between sonographic and biochemical markers can only be captured by models with high representational capacity. Using a large clinical dataset, we demonstrate the hierarchical superiority of a Deep Neural Network (DNN), which achieved a test set Area Under the Curve (AUC) of 0.982 and a Matthews Correlation Coefficient (MCC) of 0.869, through the application of SHapley Additive exPlanations (SHAP), we deconstruct the model to reveal that higher-order interaction effects account for approximately 20% of its predictive power. Furthermore, by employing a Bayesian Neural Network (BNN), we introduce a framework for quantifying predictive uncertainty, decomposing it into its aleatoric and epistemic components, the results show that the BNN can reliably flag atypical patient profiles by exhibiting high epistemic uncertainty, a critical feature for clinical safety, this work concludes that the problem's underlying geometry is that of complex, intertwined manifolds, and that models capable of learning these structures while quantifying their own uncertainty represent the next frontier in prenatal diagnostics.