Publication Search

This research benchmarks multiple machine learning (ML) algorithms for large-scale loan default prediction using a real-world dataset of 255,000 borrower records, where default cases represent only ~9–12% of total observations. The study addresses the persistent gap in comparative analyses of ML models that balance predictive accuracy, interpretability, and computational efficiency for credit risk assessment. Six algorithmic families were evaluated Logistic Regression, Random Forest, XGBoost, LightGBM, CatBoost, Artificial Neural Networks (ANN), and Stacked Ensemble—using standardized preprocessing, hybrid imbalance handling (SMOTE, class weighting, under-sampling), and comprehensive evaluation metrics (AUC, F1, Recall, Precision, PR-AUC, and Brier Score). Empirical results show Logistic Regression achieved the highest AUC of 0.732, outperforming nonlinear models under the baseline configuration, while LightGBM attained perfect recall (1.0) but low precision (0.116), indicating over-prediction of defaults. Gradient boosting models demonstrated robust calibration (Brier ≈ 0.114–0.116) and the best computational efficiency, with LightGBM showing the fastest training and lowest memory use. CatBoost exhibited strong recall but the slowest computation, and ANN underperformed on tabular data (AUC ≈ 0.56). The Stacked Ensemble delivered balanced results with AUC = 0.664 and improved overall stability. These findings confirm that boosting-based models, particularly LightGBM and CatBoost, offer superior scalability and calibration, whereas Logistic Regression remains a valuable interpretable baseline. The study concludes that effective default prediction requires integrating rebalancing, calibration, and threshold optimization to enhance recall and operational deployment reliability in large-scale credit ecosystems.

The integration of Artificial Intelligence (AI) into precision agriculture has significantly improved plant disease recognition; however, many existing deep learning models remain computationally expensive and feature-redundant, limiting their deployment on low-power and edge devices. To address these limitations, this study proposes a lightweight framework for maize leaf disease recognition based on serial deep feature extraction, dimensionality reduction, and machine-learning–based classification. A pre-trained MobileNetV2 network is employed as a fixed feature extractor to obtain discriminative visual representations, while Principal Component Analysis (PCA) is applied to reduce feature dimensionality by approximately 76%, retaining 95% of the original variance and improving computational efficiency. The compressed features are subsequently classified using a Radial Basis Function Support Vector Machine (RBF-SVM), optimized via grid search and cross-validation. Experiments conducted on a four-class maize leaf disease dataset (Northern Leaf Blight, Common Rust, Gray Leaf Spot, and Healthy), with class imbalance handled during training, demonstrate that the proposed MobileNetV2–PCA–SVM pipeline achieves 97.58% accuracy, 96.60% precision, 96.59% recall, and 96.59% F1-score, outperforming the DenseNet201 + Bayesian-optimized SVM baseline (94.60%, 94.40%, 94.40%, and 94.40%, respectively). This improvement corresponds to a 2.98% accuracy gain, a 55% reduction in error rate, an 86% reduction in model parameters (20.31M to 2.75M), and an 85% reduction in model size (81 MB to 12 MB). These results indicate that the proposed framework provides a compact and efficient solution with strong potential for deployment in resource-constrained agricultural environments.

The increasing adoption of collaborative robots in modern manufacturing environments requires reliable perception systems that can ensure both safety and operational efficiency during human–robot collaboration. This study proposes a CNN-based real-time computer vision system for object and human detection in shared robotic workspaces. The research focuses on developing and evaluating a single-stage deep learning detection model optimized for real-time performance while maintaining high detection accuracy. The proposed methodology includes dataset preparation, model training using transfer learning, real-time system implementation, and comprehensive performance evaluation. Experimental results demonstrate that the developed system achieves high detection accuracy, as reflected by strong precision, recall, and mean Average Precision (mAP) values, while maintaining low inference latency suitable for real-time operation. The system consistently operates above real-time frame-rate thresholds, ensuring timely perception updates required for safety-related decision-making in collaborative robotic environments. Graphical and quantitative analyses further confirm the stability of inference performance under dynamic interaction scenarios involving human movement and multiple objects. Compared with existing approaches, the proposed system provides a balanced trade-off between accuracy and computational efficiency, making it practical for deployment in safety-aware human–robot collaboration scenarios. Overall, the findings indicate that CNN-based real-time object detection systems can effectively support perception and situational awareness in collaborative robotics, contributing to safer and more efficient industrial automation.

Concept drift, the phenomenon where the statistical properties of data streams change over time, poses a significant challenge in machine learning, particularly for long term data streams. Traditional machine learning models, including batch learning and non-adaptive approaches, struggle to detect and adapt to these changes, leading to degraded performance and inaccurate predictions. This study proposes an adaptive computational model designed to detect and respond to concept drift using incremental learning techniques and statistical drift detection mechanisms. The model integrates an Adaptive Drift Detector (ADD) and Incremental Learning System, enabling real-time adjustments to data distribution changes. The model is evaluated across synthetic and real-world datasets, demonstrating its superior ability to detect abrupt, gradual, and recurring drifts compared to traditional models. Experimental results indicate that the adaptive model maintains high prediction accuracy, minimizes false positive rates, and reduces detection delays. Furthermore, the model performs well in resource-constrained environments, making it suitable for real-time applications such as healthcare prediction, fault detection, and IoT systems. Despite its promising performance, the study identifies challenges related to computational complexity and the model’s performance with imbalanced datasets and noisy data. Future research should focus on optimizing the model’s scalability, computational efficiency, and adaptability to more complex data types to ensure broader applicability in dynamic environments. This work contributes to advancing the detection and adaptation of concept drift, offering a robust solution for dynamic and evolving data streams.

The rapid proliferation of Internet of Things (IoT) devices across diverse industries has significantly increased the vulnerability of IoT edge networks to sophisticated cyber threats. Traditional intrusion detection systems (IDS), such as signature-based and anomaly-based approaches, are often insufficient in addressing the dynamic and evolving nature of these threats. This study proposes a hybrid intrusion detection system (IDS) framework that combines supervised machine learning (ML) techniques with deep reinforcement learning (DRL) to enhance detection performance in real-time, resource-constrained IoT environments. The proposed framework utilizes supervised learning for initial traffic classification and DRL for adaptive decision-making, enabling the system to continuously learn and optimize its detection policies based on new attack patterns. The hybrid approach significantly improves detection accuracy and reduces false positives when compared to conventional signature-based and single-model ML systems. In addition to improved detection capabilities, the framework's computational efficiency allows it to operate effectively within the constraints of IoT devices, ensuring that it is suitable for large-scale deployments. Benchmark evaluations using publicly available datasets, such as NSL-KDD, IoT-23, and BoT-IoT, show that the hybrid IDS framework outperforms traditional methods, providing a more robust and adaptive solution to cybersecurity challenges in IoT edge networks. The findings of this study suggest that combining machine learning with deep reinforcement learning offers a promising approach to secure IoT environments and address the limitations of existing IDS techniques. Future work will explore enhancing real-time adaptability, scalability, and the detection of zero-day attacks in evolving IoT ecosystems.

This research focuses on the design and evaluation of a novel parallel graph optimization algorithm incorporating dynamic load balancing (DLB) to address inefficiencies in heterogeneous computing environments. Large-scale graph optimization problems, such as those in social networks, bioinformatics, and transportation systems, often suffer from computational imbalances when using traditional static load balancing approaches, leading to underutilized resources and prolonged execution times. The primary objective of this research is to develop an algorithm that can dynamically adjust workload distribution across processors, enhancing computational efficiency and scalability. The proposed method combines heuristic techniques, including region expansion and multilevel partitioning, with diffusive load balancing strategies to minimize inter-processor communication overhead. Experimental results demonstrate that the proposed algorithm reduces execution time by up to 40% compared to static methods, with optimized resource utilization and more balanced workload distribution. The scalability of the algorithm is also evident, as it adapts effectively to increasing problem sizes and processor counts. These findings suggest that dynamic load balancing is crucial for improving parallel graph optimization in real-world applications. Future work will focus on further enhancing the algorithm’s responsiveness to rapidly changing workloads and expanding its applicability to additional domains.

Information system security faces serious challenges due to increasingly complex cyber attacks. Intrusion Detection Systems (IDS) require efficient approaches to handle high-dimensional data such as the NSL-KDD dataset with 41 features. This study aims to implement the Genetic Algorithm (GA) for feature selection on the NSL-KDD dataset to improve the efficiency and accuracy of network attack detection. The method used is computational experimental research, involving data preprocessing, GA implementation for feature selection, building a classification model using Random Forest, and performance evaluation based on accuracy, precision, recall, F1-score, and computation time. The results show that GA successfully reduced features from 41 to 12 features (70.7% reduction), significantly improving computational efficiency. However, model accuracy slightly decreased from 0.4973 to 0.4951, indicating that while GA is effective for feature selection, the elimination of certain features may reduce classification capability. The implication of this study is that GA can be used as a tool to simplify intrusion detection models, but it should be combined with parameter optimization and data imbalance handling to achieve more optimal performance.  

This research explores the effectiveness of heuristic techniques for solving combinatorial optimization problems, with a particular focus on the Traveling Salesman Problem (TSP). Combinatorial optimization is a critical area of study, especially in fields like computer science, engineering, and economics, where finding optimal solutions from a finite set of possibilities is crucial. However, the NP-hard nature of many combinatorial problems, such as the TSP, makes traditional exact methods like Branch-and-Bound and Dynamic Programming computationally expensive and inefficient for larger problem sizes. The primary objective of this research is to evaluate the performance of heuristic methods, including Simulated Annealing (SA), Genetic Algorithms (GA), and Iterative Computation techniques, such as Tabu Search (TS) and Particle Swarm Optimization (PSO). These methods are tested for their ability to provide approximate solutions efficiently. The findings reveal that while ACO provided the best solution quality, it had the longest runtime. TS was the fastest, though with slightly lower solution quality. SA and GA demonstrated a balance between solution quality and computational efficiency, but their performance heavily depended on parameter tuning. The hybridization of SA and GA showed potential for improving solution quality but introduced additional complexity. The research concludes that heuristic methods, especially when combined, offer viable solutions for large-scale combinatorial optimization problems, though the trade-off between solution quality and computational time must be considered when selecting an algorithm.

Integral equations are essential tools in applied mathematics, with wide-ranging applications in fields such as physics, engineering, and finance. However, solving these equations presents significant challenges, particularly when dealing with complex, high-dimensional, or singular problems. Traditional methods, such as manual analytical techniques or direct numerical approaches, often struggle with computational efficiency, especially for large-scale systems, and may not be suitable for handling ill-conditioned problems. This study aims to develop an efficient numerical method for solving integral equations by combining adaptive quadrature techniques with Python-based iterative solvers. The adaptive quadrature method adjusts the step size dynamically based on error estimates, ensuring high accuracy even in the presence of singularities or near-singularities, which are common in many real-world problems. The iterative solver, based on Krylov subspace methods, enhances computational efficiency by reducing memory usage and improving the convergence speed of the solution. By using these techniques together, the proposed method significantly improves the computational time required to solve large-scale and complex systems of integral equations, while maintaining satisfactory accuracy. The results demonstrate that the adaptive quadrature technique, when combined with the Python-based iterative solver, offers a substantial advantage in both speed and precision compared to traditional methods. The proposed method is especially effective in handling complex, high-dimensional systems and ill-conditioned problems, making it a powerful tool for applied mathematics, physics, and engineering applications. In conclusion, this study presents a robust and efficient approach for solving integral equations, with potential for future research in solving non-linear and multi-dimensional integral equations.

Phishing is an evolving form of cybercrime that targets users' sensitive information through URL manipulation. Conventional detection methods such as blacklists and signature-based approaches have become increasingly inadequate in addressing the dynamic variations of modern phishing attacks. This study evaluates the effectiveness of Recurrent Neural Network (RNN) variants, such Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), and Gated Recurrent Unit (GRU), in detecting phishing threats based on URL data. The methodology involves a Systematic Literature Review (SLR) of scholarly publications from the past ten years, complemented by experimental implementation of the models using a public dataset from Kaggle. Literature findings show that Bi-LSTM consistently achieves the highest accuracy, up to 99%, while GRU stands out for its computational efficiency. Experimental results support these findings, with Bi-LSTM achieving an accuracy of 96.22%, GRU 96.29%, and LSTM 95.43%. Classification metrics indicate that RNN-based models perform very well in detecting benign and defacement URLs, although their performance on phishing URLs remains challenged, particularly in terms of recall. These results confirm that RNNs remain a promising approach for phishing detection systems, especially when integrated into hybrid models with complementary architectures. This study is expected to provide a foundation for developing precise and adaptive AI systems to combat increasingly sophisticated phishing threats.

This study investigates the solutions of electrical current in linear circuit systems using two numerical methods implemented in MATLAB: the Matrix Inverse Method and the Gauss-Jordan Elimination Method. The objective is to analyze the effectiveness, accuracy, and computational efficiency of both techniques in solving systems of linear equations derived from Kirchhoff's laws. Several circuit models with varying levels of complexity are tested to compare results obtained from each method. The findings indicate that both methods yield consistent solutions, although differences in computational steps and processing time are observed. This research highlights the practicality of MATLAB as a powerful tool for electrical circuit analysis and provides insights into the selection of appropriate numerical methods for solving engineering problems.

Nonlinear eigenvalue problems (NEPs) pose significant challenges in mathematical physics and other computational applications due to their nonlinear nature, which makes analytical solutions difficult to obtain. NEPs are encountered in various scientific and engineering fields, including signal processing, electronic structure calculations, and structural optimization. This study aims to explore the application of adaptive algorithms in solving nonlinear eigenvalue problems, with a primary focus on improving accuracy and computational efficiency. The proposed method combines an iterative solver with adaptive step-size adjustment, where the step size is dynamically adjusted during the iteration based on error estimates calculated at each step. This approach enables faster convergence and significant reductions in computational time without compromising accuracy. In experiments conducted on large-scale problems, the adaptive algorithm reduced computational time by 40% faster compared to fixed-step iterative methods. The comparison between the adaptive algorithm and traditional methods showed that the adaptive algorithm is not only more efficient but also more robust when dealing with high-complexity problems. Additionally, the adaptive algorithm provides more accurate error estimates, allowing better error control throughout the iteration process. Overall, this study concludes that adaptive algorithms offer a more effective and efficient solution for complex nonlinear eigenvalue problems and can be adapted to various types of problems in scientific and engineering applications. Further research could focus on optimizing the implementation of this algorithm for larger and more complex scales.

Health risk stratification is crucial for preventive healthcare, yet existing models often rely on binary classification generalized disease prediction, neglecting personalized health indicators and graded risk levels. Many studies apply feature selection techniques like Relief and Univariate Selection without quantifying the weighted impact of features. To address these gaps, this study introduces a Big Data-driven Health Index (HI) framework using PySpark for scalable health risk stratification. The HI is computed as a weighted sum of health-related features using SHAP Analysis, XGBoost, Random Forest, and Correlation Analysis. PySpark enables efficient processing of large-scale health data, and individuals are classified into Low and High Risk. Optimal classification thresholds are determined using the Youden Index from the ROC curve to balance sensitivity and specificity. Personalized health recommendations are generated based on risk categories to guide preventive interventions. Performance evaluation reveals that Correlation Analysis achieves 100% precision and 98.90% recall, outperforming other methods. SHAP prioritizes recall but has low precision, while XGBoost and Random Forest improve precision but struggle with recall. By leveraging Big Data techniques with PySpark, this study enhances computational efficiency, scalability, and classification accuracy, addressing prior research limitations and providing a robust data-driven approach to personalized health monitoring.

This study enhances Jiang et al.'s compression-based classification algorithm by addressing its limitations in detecting semantic similarities between text documents. The proposed improvements focus on unigram extraction and optimized concatenation, eliminating reliance on entire document compression. By compressing extracted unigrams, the algorithm mitigates sliding window limitations inherent to gzip, improving compression efficiency and similarity detection. The optimized concatenation strategy replaces direct concatenation with the union of unigrams, reducing redundancy and enhancing the accuracy of Normalized Compression Distance (NCD) calculations. Experimental results across datasets of varying sizes and complexities demonstrate an average accuracy improvement of 5.73%, with gains of up to 11% on datasets containing longer documents. Notably, these improvements are more pronounced in datasets with high-label diversity and complex text structures. The methodology achieves these results while maintaining computational efficiency, making it suitable for resource-constrained environments. This study provides a robust, scalable solution for text classification, emphasizing lightweight preprocessing techniques to achieve efficient compression, which in turn enables more accurate classification.

Combinatorial optimization is a fundamental area in operations research and computer science, focusing on identifying optimal solutions from a finite set of possibilities. This study explores the integration of branch and bound methods with heuristic algorithms to address optimization problems in graph theory and discrete mathematics. Python was employed for algorithm implementation due to its flexibility and comprehensive computational libraries, enabling efficient data analysis and visualization. Several benchmark problems were examined, including the Traveling Salesman Problem (TSP), Minimum Spanning Tree (MST), and Graph Coloring. Simulations were conducted using datasets of varying sizes (small, medium, and large) to evaluate performance across different scales. The results demonstrate that the hybrid approach achieves a balance between solution quality and computational efficiency, outperforming brute-force methods in terms of speed while maintaining near-optimal accuracy. Tabulated results and graphical comparisons highlight the reduction in computation time and improved scalability of the proposed method. The findings suggest that combining systematic search strategies with heuristics offers a practical and effective framework for solving complex combinatorial optimization problems. Recommendations for future research include testing scalability with larger datasets, incorporating advanced metaheuristics, and applying the approach to real-world domains such as logistics and network design.

This research explores the use of iterative methods in conjunction with the Finite Difference Method (FDM) for solving partial differential equations (PDE). The central challenge addressed is the computational inefficiency and slow convergence that often arise when utilizing traditional numerical methods, particularly in large-scale systems. The study aims to develop a more efficient iterative approach to solve PDEs by minimizing computational time while ensuring the stability of the obtained solutions. The primary methods proposed include iterative solvers such as Gauss-Seidel and Successive Over-Relaxation (SOR), which are applied to numerical solutions derived from FDM. The research demonstrates that iterative methods, especially SOR, achieve faster convergence with fewer iterations compared to conventional methods like the Finite Element Method (FEM), which tends to be slower and more resource-intensive for large-scale problems. The study highlights the advantages of iterative solvers in efficiently handling large, sparse linear systems and reducing computational costs. In addition, it shows that these methods are capable of providing stable solutions, thereby maintaining accuracy with significantly lower computational effort. The results suggest that iterative methods, when applied in combination with FDM, offer a practical and scalable solution for solving complex PDEs. These methods are especially beneficial in engineering and theoretical physics applications where large-scale simulations are prevalent. The study concludes with recommendations for future research, which should focus on further optimizing solver parameters, exploring hybrid approaches, and extending the methods to more complex PDEs with non-linearities or irregular geometries. By doing so, these techniques could contribute to even more efficient and practical solutions for real-world applications.

This paper explores the Time-Based One-Time Password (TOTP) authentication mechanism enhanced with lightweight cryptographic algorithms, presenting it as an advanced solution to the limitations of traditional OTP systems. There are a lot of applications and systems where this mechanism is applied. For example, bank applications, e-commerce websites, access control system, healthcare system, etc. TOTP generates dynamic, time-sensitive passwords using the current time and a secret key processed through a cryptographic hash function, significantly improving security by reducing vulnerabilities to code reused and interception. The adoption of lightweight algorithms ensures that TOTP can be efficiently implemented on resource-constrained devices, such as those on the Internet of Things (IoT) ecosystem. Despite its benefits, TOTP faces challenges including synchronization issues between client devices and servers, and a trade-off between computational efficiency and security strength. This paper discusses the implications of these challenges and evaluates how TOTP, with appropriate design considerations, can provide a robust, secure, and efficient authentication method suitable for various applications, from digital banking to IoT environments.

Big data has revolutionized the landscape of natural sciences by providing extensive datasets that enable deeper insights and more accurate predictions. However, effectively analyzing such vast and complex data requires optimized machine learning algorithms tailored to specific applications. This study focuses on enhancing the performance of machine learning models in big data analysis for applications in natural sciences. The research aims to identify key optimization techniques, including feature selection, hyperparameter tuning, and algorithm customization, to improve model accuracy and computational efficiency. A combination of supervised and unsupervised learning approaches was applied to real-world datasets in fields such as climate science, genomics, and ecology. The findings demonstrate significant improvements in predictive accuracy and processing speed, highlighting the potential of optimized machine learning techniques in solving complex problems in natural sciences. The implications of this research extend to more efficient resource utilization and improved decision-making in scientific exploration and environmental management.

Quantum computing offers promising alternatives to classical approaches for solving complex linear algebra problems. This paper presents a comparative study of the performance of quantum algorithms versus classical algorithms in solving systems of linear equations and matrix operations. Through simulation and analysis, we demonstrate that while quantum computing holds advantages in specific problem sets, classical computing remains efficient for general applications. These findings highlight the current limitations and potential of quantum computing.

The rapid growth of data-intensive applications has posed significant challenges for classical machine learning (ML) algorithms, particularly in terms of computational efficiency and scalability. This study explores the role of quantum computing in optimizing machine learning performance through the implementation of Quantum Machine Learning (QML), specifically using the Quantum Support Vector Machine (QSVM) model. The research adopts a Design Science Research approach, involving problem identification, model development, system implementation, and performance evaluation. Both classical Support Vector Machine (SVM) and QSVM models are developed and tested using benchmark classification datasets. The results indicate that QSVM outperforms the classical SVM model across multiple evaluation metrics, including accuracy, precision, recall, and F1-score. Additionally, QSVM demonstrates improved computational efficiency by reducing training time, particularly when handling high-dimensional data. These improvements are attributed to the ability of quantum computing to utilize quantum kernel methods and map data into higher-dimensional feature spaces, enabling better pattern recognition and classification performance.  Despite these promising outcomes, the study also identifies several limitations related to current quantum hardware, such as noise, decoherence, and limited qubit availability, which may affect scalability and practical implementation. Therefore, further research is required to enhance quantum hardware reliability and develop hybrid quantum-classical models. In conclusion, quantum machine learning offers a promising solution to overcome the limitations of classical approaches, providing enhanced performance and efficiency for complex data processing tasks in future intelligent systems.