This study evaluates a common experimental framework comprising deep learning, ensemble learning, and optimization-based feature selection techniques for spam email detection. To date, knowledge-based and machine learning-based methods have been used for spam detection. However, classical machine learning-based methods can be insufficient in representing and classifying complex spam content. In this study, experiments were conducted on the Enron and SpamAssassin datasets using CNN, GRU, XGBoost, and Random Forest algorithms, a genetic algorithm-based feature selection approach, and TF-IDF, Word2Vec, and embedding-layer representations. The contribution is the cross-comparison of established model, representation, and genetic-algorithm feature-selection combinations on two benchmark datasets rather than a new learning algorithm. Experimental results are presented using metrics such as AUC, F1 Score, and accuracy. The maximum F1-score and AUC value obtained in our study were 0.9997.
Asymmetric oscillations occur in numerous physical systems, where the restoring force is often characterized by non-polynomial stiffness nonlinearities. In this study, an equivalent Helmholtz–Duffing oscillator was developed to derive approximate analytical solutions for the periodic response of asymmetric oscillators with non-polynomial stiffness nonlinearities under admissible general initial conditions. The equivalent oscillator was formulated using quasi-static equilibrium principles, resulting in amplitude-dependent stiffness coefficients. The contribution is the mixed-parity force-and-energy matching construction, rather than a new quasi-static principle or a new exact solution technique. The exact closed-form solution of the equivalent Helmholtz–Duffing oscillator was then employed to obtain an approximate analytical solution for the original asymmetric system. The reported comparisons indicate lower period errors than the specified cubic Taylor approximations in the selected moderate and strongly nonlinear cases. Unreconciled numerical and normalization details limit this validation; a general accuracy or computational-efficiency advantage is not established.
This work presents a nonlinear dynamic analysis and delayed proportional–derivative (PD) control of piles subjected to prescribed lateral excitations. The model incorporates geometric nonlinearity, nonlinear Winkler-type soil reactions, and modal coupling within a Lagrangian framework. The nonlinear partial differential equation is examined through modal projection and through a separate semi-discretized finite-difference formulation integrated with a Runge–Kutta scheme. The reported numerical results exhibit approximately periodic, amplitude-modulated, dissipative, bursting-like, and irregular finite-time responses as the excitation frequency, load amplitude, and soil stiffness are varied. Harmonic excitation is treated as additive forcing, whereas constant-load responses are interpreted as transients rather than as evidence of sustained chaotic motion. The PDE simulations illustrate mixed oscillations, convergence toward a steady displacement under constant loading, and asymmetric bending distributions along the pile depth. A delayed PD controller is examined in modal space, and residual forced oscillations remain visible in the reported controlled response. The stability analysis concerns the unforced linearized modes and therefore does not establish nonlinear robustness or exact disturbance rejection. The contribution is an application-oriented study based on established modeling and control techniques. The three-oscillator approximation and the clamped–free PDE calculations employ different boundary idealizations; consequently, matched-model validation, physical calibration, and complete reproducible simulation details remain necessary before predictive engineering use.
Ethiopia’s construction sector is rapidly expanding and growing more technically and managerially complex, especially in public building projects, yet many small- and medium-sized domestic contractors still fail to meet cost, schedule, and quality targets. This study examines the stakeholder-rated factors relating to project performance, identifies key performance indicators, and evaluates current performance measurement practices in public building construction. A mixed-methods design combining qualitative and quantitative approaches was used. Data was collected through questionnaires and case studies and analyzed with the Relative Importance Index and SPSS (version 20). The study identified perceived performance constraints: delayed payments, rising material costs, exchange rate volatility, design changes, labor relations, availability of highly qualified personnel, climatic conditions, and project managers’ leadership effectiveness. These factors were rated in relation to construction project time, cost, quality, health and safety, client satisfaction, and productivity; the ratings do not establish their causal effects. Results show that contractors reported using formal, quarterly performance measurement systems and earned value analysis. The contribution is a Gondar-specific descriptive prioritization and conceptual framework, not a validated performance-improvement method.
Facial recognition systems often experience substantial performance degradation when essential facial regions are obscured by occlusions. These occlusions interfere with feature visibility and reduce the accuracy and robustness of conventional single-stage training processes. To address this challenge, this study proposes a multi-stage training framework designed to improve facial recognition robustness under occluded conditions. The framework consists of three progressive training phases: initial feature learning, occlusion-specific adaptation, and refinement and fine-tuning, with each phase guided by a specific loss configuration and optimization strategy. Targeted data augmentation techniques were incorporated to improve generalization under different occlusion conditions. Experiments were conducted using 12,000 facial images compiled from Labeled Faces in the Wild (LFW), CelebFaces Attributes Dataset (CelebA), and a MAFA masked-face dataset subset, with synthetic occlusions introduced for occlusion-focused training and evaluation. The proposed model achieved a final recognition accuracy of 89.7%, compared with 78.9% for the conventional single-stage CNN, representing a 10.8 percentage-point improvement. Convergence was approximately 25% faster than the baseline. Across face-mask, sunglasses, scarf, and combined-occlusion categories, the multi-stage model consistently achieved higher recognition accuracy than the single-stage model. Precision, recall, and F1-score were also reported for the multi-stage model, whereas the corresponding category-level baseline values were not retained in the available experimental record. These findings indicate that progressive training with occlusion-specific adaptation can improve facial recognition robustness under partially obscured conditions, although the additional computational requirements should be considered for resource-constrained deployment.
Environmental chemistry has entered a data-rich era in which analytical instruments, sensor networks, and remote-sensing platforms generate enormous volumes of complex information. Statistical methods have long been the foundation of environmental analysis, offering hypothesis testing, regression, variance analysis, and multivariate approaches for exploring patterns in pollutant distribution and chemical interactions. While these classical approaches remain essential for experimental design, calibration, and uncertainty assessment, the emergence of machine learning has provided new opportunities to address the high dimensionality, nonlinearity, and heterogeneity of modern environmental datasets. From random forests and support vector machines to gradient boosting and deep neural networks, these tools have been applied to problems such as pollutant source apportionment, non-target screening in high-resolution mass spectrometry, spectral deconvolution, sensor fusion, and water-quality forecasting. Their capacity to recognize complex patterns can deliver predictive accuracy that surpasses traditional methods in many study-specific comparisons, yet their adoption has introduced new concerns related to overfitting, data leakage, lack of interpretability, and poor reproducibility. In order to fully harness the benefits of machine learning in environmental chemistry, it is critical to embed domain knowledge, validate models using robust statistical frameworks, quantify uncertainty, and establish standardized practices for transparency and reproducibility. This review examines the evolution of statistical and machine learning tools in environmental chemistry, analyzes their applications and comparative strengths, highlights challenges and pitfalls, and proposes future directions that emphasize hybrid physics–ML models, interpretability, reproducibility, and ethical use. By integrating statistical rigor with computational innovation, environmental chemistry can better support discovery, decision-making, and sustainable management of pollutants.
The influence of combinations of granulated blast furnace slag (GBFS), pulverized fly ash (PFA), and silica fume (SF) on the properties of ordinary Portland cement (OPC) pastes was investigated. The incorporation of GBFS, PFA, and SF increased the standard water of consistency, while the initial and final setting times generally followed the same tendency. Chemically combined water content and bulk density increased with curing time, whereas the free-lime content of the blended pastes and the total porosity decreased. Both flexural and compressive strengths of the hardened cement pastes increased with curing time. At the reported curing ages, the M3 and M4 blends exceeded the OPC control, whereas M1 and M2 remained below it. Among the formulations examined, the composite containing equal proportions of the three industrial by-products (M4) exhibited the most favorable overall response across the measured fresh, hydration, physical, mechanical, thermal, and microstructural properties. DTA–TG analysis indicated a greater amount of C–S–H-related hydrates and a lower amount of Ca(OH)\(_2\) in the blended systems than in the OPC paste. These findings show that the combined use of GBFS, PFA, and SF at the investigated total replacement level can improve several characteristics of Portland cement pastes while simultaneously promoting the beneficial utilization of industrial by-products.
This paper examines how logical structures represent sharp and unsharp propositions in classical and quantum systems. Classical sharp logic is modeled by Boolean algebras, where propositions have bivalent truth values, complements are globally defined, and distributivity holds. Quantum sharp logic is modeled by orthomodular lattices of closed Hilbert-space subspaces or projection operators; it retains definite yes–no outcomes for ideal measurements but rejects the Boolean assumptions of global truth valuation and distributivity. Classical unsharp logic uses fuzzy membership or probabilistic truth values to describe vagueness and incomplete information, whereas quantum unsharp logic is expressed through effects and effect algebras, which capture noisy, inefficient, and generalized measurements. The paper’s central question is which algebraic structure is appropriate for each combination of classical or quantum behavior and sharp or unsharp measurement. The analysis shows that the four cases form a coherent hierarchy: Boolean algebras describe deterministic classical events, orthomodular lattices describe ideal quantum propositions, fuzzy or probabilistic models describe classical partial truth, and effect algebras describe realistic quantum effects. This classification clarifies the mathematical and physical role of unsharpness and explains why effect algebras are needed beyond projection-based quantum logic.
Reliable rainfall–runoff estimation is essential for the hydraulic design and planning of canal systems, particularly in ungauged catchments where streamflow records are unavailable. This study aimed to generate design-storm hydrographs and estimate peak runoff for the Narai Canal catchment in Peshawar, Khyber Pakhtunkhwa, Pakistan, using the Log-Pearson Type III (LP-III) distribution and the WinTR-20 model under ungauged-basin conditions. Rainfall frequency analysis was performed using the LP-III distribution to estimate design rainfall for selected return periods. The Soil Conservation Service (SCS) curve number method implemented in WinTR-20 was then used to simulate runoff hydrographs and peak discharges based on watershed characteristics, including land use, curve number, and time of concentration. The analysis produced design-storm hydrographs and corresponding peak discharge estimates that increased with rainfall intensity and return period. The LP-III distribution adequately characterized design rainfall, while the WinTR-20 model provided physically consistent runoff estimates for evaluating watershed response under different design scenarios. The results demonstrate the influence of watershed characteristics on runoff generation and provide preliminary hydrological estimates for the study area, representing a design-scenario analysis for an ungauged basin rather than a validated rainfall–runoff forecasting model. The generated design hydrographs and peak discharge estimates provide useful information for preliminary canal hydraulic design, drainage planning, and flood-risk assessment. Future studies should incorporate observed streamflow data to calibrate and validate the model before it is applied for operational flood forecasting.
