Search Results (1,695)

Continuous streamflow monitoring is fundamental for water management in high-mountain Andean basins. Traditionally, this process relies on empirical regressions, although artificial intelligence (AI) has recently emerged as a robust alternative. However, extreme geomorphological dynamics compromise classical hydraulic methods, while AI models frequently lack physical validation. In this context, this study compares the performance of Artificial Neural Networks against traditional methods to reduce uncertainty in stage–discharge rating curves. The methodology, applied to a nested basin scheme in Loja, Ecuador, contrasted traditional exponential fits with a Multilayer Perceptron optimized using the Levenberg–Marquardt algorithm. The analysis included the evaluation of uncertainty bands and a sub-hourly spatial validation based on the principle of mass conservation. Results evidence that AI refines statistical accuracy (NSE > 0.95) and effectively adapts to bed non-linearity; nevertheless, cross-validation revealed a high susceptibility to algorithmic overfitting. It is concluded that while AI offers superior analytical flexibility for interpolating non-linear dynamics, traditional methods remain more robust for extreme flood extrapolation. Furthermore, while AI reduces computational complexity, it entails a higher “data cost” requiring denser field gauging campaigns. Operational viability requires rigorous dynamic uncertainty controls and spatial water balance validation. Full article

Due to the complex issues of low porosity and low permeability in tight sandstone reservoirs, non-unified data measurement, and the limitation of traditional methods by empirical formulas and simple statistical models, which make it difficult to couple the correlation of parameters, how to quickly clean data, establish a comprehensive geological-engineering sweet spot evaluation method, and improve prediction accuracy and engineering decision-making effectiveness have become an urgent technical challenge. This study takes the logging and fracturing construction data in the L area as the data set, uses the Pearson correlation coefficient method to verify the nonlinear characteristics of features, and constructs a geological-engineering integrated intelligent decision-making algorithm based on the collaborative optimization of a dual-branch multi-layer perceptron and attention mechanism. The training results of the dual-branch multi-layer perceptron model and traditional machine learning methods are compared and analyzed. The results show that the prediction error of the adopted dual-branch multi-layer perceptron neural network model is 5.44%. The weight of geological factors in this area accounts for 51.71%, and the engineering factors account for 48.29%. This method has been field-applied in 25 wells in the L area, with a production coincidence rate reaching 94.66%. The sweet spots of tight sandstone reservoirs are mainly the H5 and H6 submembers. The deep integration of machine learning interpretability and geological engineering practice provides a new approach for sweet spot prediction. Full article

This research aimed to develop a rapid, non-destructive, and accurate method for detecting adulteration in turmeric using Visible–Near-Infrared (UV/Vis and NIR) spectroscopy combined with machine learning algorithms. Spectral data from the samples were collected and analyzed in two ranges: 170–870 nm (UV/Vis) and 900–2170 nm (NIR). Four supervised learning algorithms, including Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), the Multilayer Perceptron (MLP) neural network, and Decision Tree, were evaluated for modeling. To quantitatively assess model performance, we employed not only the accuracy metric but also complementary performance indicators including precision, recall, and the F1-score to provide a more comprehensive evaluation of classification effectiveness. The models developed in the 900–2170 nm spectral range demonstrated highly significant performance, with most models achieving 100% accuracy on the independent test set. To reduce data dimensionality and enhance computational efficiency, a hybrid feature selection method combining SVM with five algorithms—League Championship Algorithm (LCA), Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), and Imperialist Competitive Algorithm (ICA)—was employed. Upon evaluation of each method, the SVM-LCA was selected as the optimal feature selection technique. This algorithm successfully extracted the most effective wavelengths with the highest correlation and lowest error, which maintained or improved the accuracy of the classification models. This study confirms the high potential of UV/Vis and NIR spectroscopy as rapid, non-destructive, and precise tools for detecting adulteration in turmeric. The findings can pave the way for the development of intelligent quality control systems in the food and pharmaceutical industries, playing a crucial role in ensuring consumer health and safety. Full article

Existing anomalous traffic detection methods based on feature fusion in Software-Defined Networking (SDN) lack adaptability in weight allocation mechanisms. Consequently, their detection accuracy and model generalization capabilities fail to meet practical security requirements. To solve these limitations, this paper proposes a refined detection method based on hybrid feature selection and gated fusion. First, the framework employs XGBoost combined with the Recursive Feature Elimination (RFE) algorithm. This process identifies shallow statistical features with high discriminative power. Simultaneously, the method utilizes a 1D Convolutional Neural Network (1D-CNN) integrated with a Squeeze-and-Excitation (SE) block to extract deep temporal semantic features. Subsequently, a tailored gated fusion mechanism incorporating linear projection layers for feature alignment adaptively integrates these two categories of features. The fused features are then input into a Multilayer Perceptron (MLP) to execute anomalous traffic detection. Experimental results demonstrate that the proposed method achieves superior performance. Specifically, on the InSDN Dataset, the binary and multi-classification accuracy rates reach 99.91% and 99.88%. Similarly, the accuracy rates on the NSL-KDD dataset are 99.78% and 99.76%. Finally, we established a local simulation environment. Experimental results demonstrate that our method attains an average precision exceeding 93% for anomalous traffic detection in simulated real scenarios. Full article

Breast cancer screening is a fundamentally probabilistic diagnostic task that requires precise identification of complex imaging characteristics from diverse patient cohorts. Despite improvements in deep learning techniques, current automatic tools are typically trained on well-curated datasets and do not generalize to heterogeneous data, thereby limiting their application. This study aims to address these shortcomings by introducing a more effective and generalizable framework for breast cancer classification that focuses on the stability of features, the learning of complementary representations, and improved decision modeling. The proposed methodology incorporates stability-driven feature extraction (SDFE) with a multi-branch architecture that consists of EfficientNetV2 (Convolutional neural networks (CNNs)), EfficientFormer (Vision transformers (ViTs)), and multi-layer perceptron (MLP)-Mixer models to extract various feature representations. To improve non-linear decision boundaries, it uses a Kolmogorov–Arnold Network (KAN)-based classification head and selects the most credible prediction via an adaptive voting mechanism. This model is trained using patient-level splitting on the VinDr-Mammo dataset, evaluated using five-fold cross-validation, and subsequently externally validated on the CBIS-DDSM dataset. Experimental findings demonstrate the consistent performance of the proposed model, with accuracies of 94.5% in cross-validation, 93.3% on the VinDr-Mammo test set, and 94.6% on CBIS-DDSM, surpassing other recent state-of-the-art solutions. It demonstrates enhanced robustness and cross-dataset generalization, offering a scalable, consistent framework for breast cancer classification that supports the development of computer-aided diagnostic systems. Full article

The transformation of healthcare systems presents a key challenge for EU Member States, as they face growing population needs, resource pressures, and ongoing health disparities among communities. Using a systems-thinking approach provides a suitable framework for analyzing how healthcare resources, behavioral and social health determinants, and overall population outcomes are interconnected. This study aims to explore these relationships from a systemic viewpoint within health communities across the European Union. The research employs a cross-sectional approach, utilizing aggregated data from all 27 EU Member States. It involves descriptive and factor analyses to create composite indices of healthcare resources, positive health determinants, and health outcomes. Furthermore, it uses a univariate Generalized Linear Model (GLM) and multilayer perceptron neural networks to model nonlinear relationships among variables. Cluster analysis is also conducted to classify Member States into different health performance typologies. The results emphasize connections between healthcare system resources and population health outcomes, illustrating how positive determinants impact health status and highlighting structural differences across European communities. The results indicate relevant associations between healthcare system resources and population health outcomes, whereas the analysis of positive health determinants suggests more complex, partly nonlinear patterns. The findings also highlight structural differences across European health communities. Full article

This paper investigates solar energy production forecasting at a monthly temporal resolution using a pooled neural network framework applied to the Chikalov photovoltaic systems in southwestern Bulgaria. The study considers several related PV installations with unequal time-series lengths and formulates the forecasting task as one-step-ahead prediction of the next monthly total energy yield, measured in kWh, in a global pooled setting. Two complementary neural architectures are compared: a multilayer perceptron (MLP), which serves as a nonlinear feed-forward benchmark based on lagged observations and seasonal descriptors, and a gated recurrent unit (GRU), which explicitly models sequential temporal dependence. In both cases, seasonality is represented through cyclical calendar encodings, while model selection is performed by chronological hyperparameter search using a separate validation block. Forecast accuracy is assessed by RMSE, MAE, coefficient of determination (R²), MAPE, and sMAPE, and uncertainty is quantified through validation residual prediction intervals. The results show that the MLP achieves stronger validation performance, whereas the GRU provides better final out-of-sample generalization after refitting on the combined training and validation data. For both architectures, the best configurations are obtained with a 12-month input horizon, indicating that one full annual cycle contains the most informative memory for forecasting monthly aggregated photovoltaic energy yield in the considered dataset. After refitting on the combined training and validation data, the GRU achieved the best final out-of-sample performance, with RMSE = 296.38 kWh, MAE = 213.16 kWh, R² = 0.9231, MAPE = 7.52%, and sMAPE = 7.49%. Overall, the findings demonstrate that pooled neural modeling is an effective framework for monthly PV production forecasting and can provide practically useful support for sustainable energy planning, monitoring, and optimization. Full article

Chloride penetration depth (CPD) is a critical durability indicator for concrete structures, yet experimental data are often limited. This study evaluates whether increasing model complexity is beneficial under such constraints by comparing six machine learning and deep learning models—extreme gradient boosting, categorical boosting (CATB), random forest, multilayer perceptron (MLP), deep neural network (DNN), and a hybrid model combined with CATB and DNN (CatDNN)—using a dataset of 1078 cases. During training, CatDNN exhibited the earliest stabilization, reaching the best epoch at 40, while MLP and DNN stabilized after approximately 30 epochs. However, overfitting tracking revealed a flat tendency near 40 epochs for CatDNN, indicating potential data saturation. The test results showed small performance differences among all the models. CatDNN achieved the lowest max error (1.21), demonstrating effective residual correction, but its R² (0.9123) was slightly lower than that of DNN (0.9129), suggesting that increased complexity did not yield meaningful improvement. The validation results confirmed high reliability across all the models (R² ≥ 0.88). Overall, the findings indicate that, under limited data conditions, simple and well-fitted models can outperform or match complex hybrid architectures, emphasizing the importance of model efficiency over structural complexity. Full article

With the large-scale grid integration of renewable energy sources such as wind power and photovoltaics, power system net load fluctuations have become significantly more severe, imposing higher demands on system flexibility. Traditional optimal transmission switching (OTS) models require the simultaneous optimization of continuous and discrete variables, resulting in high computational complexity that renders them unsuitable for daily real-time scheduling in large-scale power systems. This paper develops a flexible real-time rolling optimization scheduling model that incorporates OTS and proposes a two-stage fast solution framework based on deep neural networks (DNN). In the offline training phase, a multilayer perceptron-based DNN is trained using load and renewable generation data to rapidly and accurately predict the optimal line switching scheme. In the online application phase, the network topology predicted by the DNN transforms the original mixed-integer linear programming problem into a standard linear programming problem, substantially reducing computational complexity and solution time. Case studies on the modified IEEE 118-bus and IEEE 300-bus systems show that the proposed method achieves high prediction accuracy, reduces solution time by up to 117 times, and maintains nearly identical system operating costs to the physics-driven approach in the majority of cases. The results demonstrate that the proposed approach effectively balances computational efficiency and economic performance, verifying the practical value of optimal transmission switching in enhancing large-scale renewable energy accommodation and overall power system flexibility. Full article

This work is devoted to the investigation of an actual scientific and applied problem in the identification of exit points for early-exit DNNs based on the analysis of neural chains, which is one of the complex tasks related to the scientific and applied problems of DNN optimization, including, in particular, those based on the existing early-exit concept. The obtained computational complexity of the developed method is not limited by the latter itself, but instead, it mainly depends on the chosen algorithm for analyzing the occurrences of particular substrings (i.e., trimmed neural chains) into a defined list of strings (i.e., full neural chains). For example, in the framework of the conducted research, the Python operator “in” has been used (for this purpose), which uses an in-built optimized algorithm based on the combination of the Boyer–Moore and Horspool algorithms with a linear scalability, and computational complexity that approaches the arithmetic product of the total number of strings (i.e., full neural chains) in the array by the average length of the string in the same array. The performed practical approbation of the developed method gave positive results in decreasing the overall time for obtaining the final result of the considered DNN, as well as significantly decreasing the following timing parameters of the considered DNN: the minimal time to obtain the final result (reduced by more than 5 times); the average time to obtain the final result (reduced by ~1.4 times); and the total time spent processing all 22,500 modeling cases in total (reduced by ~1.39 times). In terms of the main positive aspects and advantages of the developed method, we could highlight its maximal versatility (in terms of the studied DNNs, their architectural and/or structural features, application areas, and input data representation, as well as further software implementation of the proposed method), together with its maximal simplicity of representation and understanding, which ensures the possibility of working with this method even for novice and inexperienced researchers and users who have only basic knowledge of DNNs. In addition, the main results and conclusions of the conducted research are given, and the prospects for further research are considered. Full article

Accurate prediction of the behavior of alloys and metal matrix composites during high-temperature deformation requires strict consideration of the loading history. To address this problem, a hybrid rheological model for flow stress prediction has been developed, combining a phenomenological description of the yield stress with a recurrent neural network based on the NARX (Nonlinear AutoRegressive with eXogenous inputs) architecture. The memory effect is formed by expanding the input parameters with the response values from the previous step. The identification of the weight coefficients of the NARX neural network is implemented by training an equivalent multilayer perceptron. To improve the generalization ability of the model and eliminate its dependence on a fixed discretization step, the training dataset includes data obtained under non-monotonic changes in the strain rate over time and a variable time interval. The article justifies the structure of the model input parameters, excluding the accumulated strain from the input set due to its lack of informativeness during active softening processes. Verification of the hybrid model on the 7075/2.5% TiC composite in the temperature range of 300–500 °C demonstrated an average relative error of 1.5% when predicting modes that were not involved in the training. The predicted flow stress values fall within the experimental scatter interval of ±5% and accurately reproduce the local features of the flow stress curves. The proposed model and its identification technique provide correct consideration of the deformation history under the complex interaction of hardening and softening processes. Full article

Background: An accurate D’Amico risk stratification is mandatory for prostate cancer (PCa) management. The purpose of this proof-of-concept study was to establish a methodological framework for integrating validated clinical nomograms with strict data-quality governance in order to generate reliable artificial neural networks (ANNs), even when the sample is small. Methods: We performed a retrospective analysis of a curated cohort of 49 patients from one centre. A multilayer perceptron (MLP) was trained using 11 variables, including the ISUP biopsy grade and Briganti nomogram. Model development was guided by a proactive data-quality protocol based on FAIR principles—the DQG-AI framework (data quality governance for AI-readiness, developed at Clínica Universidad de Navarra)—with stringent checks for accuracy, consistency and validity to ensure data were “AI-ready”. A sensitivity analysis was conducted on three data partitioning scenarios (20/80, 34/66 and 39/61). Results: From a starting pool of 76 patients, the DQG-AI framework was applied to create a highly selected cohort of 49 patients. A multilayer perceptron (MLP) trained on this “AI-ready” dataset achieved, on the 20/80 configuration, mathematically perfect discrimination (AUC 1.000; 100% accuracy) for High vs. Intermediate risk groups on a very small refined internal test set (N = 9), a figure we interpret as a methodological artefact of the curated dataset and validation constraints rather than as an indicator of true model performance. This complete accuracy is not, however, presented as evidence of generalizable clinical utility: it is a best-case figure obtained on a single, very small test subset (N = 9) after necessary validation-related exclusions, and the wide confidence interval (66.4–100%), together with the software-driven removal of test cases carrying factor levels absent from the training set (detailed in the Methods section), explicitly preclude any inference about real-world performance. Accordingly, the deliverable of this proof-of-concept study is the DQG-AI framework itself, not the model’s reported accuracy. Conclusions: The main contribution of this proof-of-concept study is the effective illustration of the DQG-AI framework as a strict, repeatable approach for producing “AI-ready” urological datasets. Although the MLP demonstrated a robust internal signal for risk discrimination, its flawless accuracy is an ideal, non-generalizable situation. The most important deliverable that needs external validation is the DQG-AI framework, not the model’s performance metrics. A pre-specified three-phase multi-institutional validation roadmap (single-centre cohort expansion → within-system between-site validation → Spanish multi-centre external validation), with a minimum target of ~220 evaluable patients derived from a 10-events-per-predictor floor, is provided to operationalise this external validation. Full article

Accurate modelling of nonlinear dynamical systems remains a fundamental challenge in control engineering, particularly in applications characterized by strong nonlinearities, uncertainty, and varying operating conditions such as anti-lock braking systems (ABSs). Although neural networks are widely used for nonlinear system identification, their performance is often limited by correlated input features, poor numerical conditioning, and reliance on computationally demanding nonlinear optimization. This paper proposes a novel neural network modelling framework that integrates improved almost-orthogonal functional input transformation with a linear-in-parameters structure. The proposed approach systematically constructs a nonlinear feature space in which correlations between basis functions are explicitly controlled through a perturbation-based near-orthogonality mechanism, resulting in improved conditioning of the regression matrix and enabling stable least-squares-based parameter estimation. The method is formulated for a general class of nonlinear discrete-time systems and experimentally validated on an Inteco ABS laboratory setup, where wheel slip dynamics are identified using measured wheel speeds and braking torque. The obtained results demonstrate improved modelling accuracy, increased robustness to measurement noise, non-Gaussian disturbances, and parameter drift, as well as lower computational complexity compared with conventional multilayer perceptron and polynomial-based models. These findings suggest that structured feature generation may improve the reliability of data-driven models and indicate potential applicability of the proposed framework for real-time and control-oriented applications in complex dynamical systems. Full article

At the Second Water Intake Pump Station of the Chenhang Reservoir in Shanghai, suboptimal pump scheduling resulted in electricity consumption cost attributable to pump-motor equipment accounting for an exceptionally large proportion of the total power expenditure. In response to the economical operation issues, a mathematical model of power consumption cost for the pump station was established by introducing time-of-use electricity pricing and constraint suppression terms. Taking the minimum cost as the research objective, the differential evolution (DE) algorithm was employed to establish a fitness function for electricity cost, aiming to find the most economical and reliable scheduling strategy. However, owing to its low computational speed and high complexity, machine learning was introduced to establish neural network surrogate models of the DE algorithm. By comparing three surrogate models, the Multilayer Perceptron (MLP) neural network model was adopted as the most appropriate surrogate model. It was optimized for robustness improvement and verified on site. The results demonstrate that implementing the surrogate model achieves over 25% savings in electricity cost per thousand cubic meters of water, while slashing the solution time by 88.53% compared to the standard DE algorithm. Furthermore, the overall power consumption is reduced by 2.20% under a cost-priority strategy and by 15.89% under a power-priority strategy, thereby directly mitigating the carbon footprint of the pump station. The proposed hybrid computational framework in this study bridges the gap between the computationally expensive heuristic optimization and the strict real-time control requirements in engineering, highlighting its significant contribution to the sustainable and low-carbon operation of water infrastructure. Full article

Processing sensitive data in cloud-based neural networks raises privacy concerns, which Homomorphic Encryption addresses by enabling privacy-preserving machine learning. In our previous work, we introduced CryptoKANs, enabling efficient Kolmogorov–Arnold Network (KAN) inference over encrypted data via polynomial approximation of spline-based activation functions using KAN symbolization. To avoid performance degradation, CryptoKAN required min–max scaling of pre-activation inputs to a small interval—a requirement that could negatively affect training. In addition, a direct theoretical structural comparison with Multi-Layer Perceptron (MLP)-based solutions, such as CryptoNets, was missing. In this work, we address these limitations by presenting CryptoKAN+, a KAN-inspired network integrating self-learned polynomial activations through a Fully Connected Quadratic Transformation (FCQT) layer. By enforcing polynomial activations during training, this design replaces spline functions without post-training symbolization, eliminates the need for interval scaling, absorbs subsequent linear transformations, and reduces multiplicative depth for efficient encrypted inference. Experiments show that CryptoKAN+ achieves competitive accuracy while slightly improving encrypted inference efficiency—a natural consequence of compacting weights with self-learned activations. Overall, this work provides a formal analysis of the structural relationship between KANs and MLPs and demonstrates how enforcing polynomial activations during training enables efficient encrypted inference while preserving accuracy. Full article