Artificial Neural Network Model Research Articles

Magnesium alloys are lightweight structural materials recognized for their exceptional strength-to-weight ratio, corrosion resistance, and superior bio compatibility, rendering them suitable for aircraft components, gearboxes, computers, mobile devices, automotive, biomedical, and electronic devices. This study investigates the modelling and prediction of abrasive waterjet drilling (AWJD) process parameters on Magnesium Alloy (AZ31B-Mg) using Artificial Neural Networks (ANN) and Grey Relational Analysis (GRA). The L9 Taguchi orthogonal array was used to examine the effects of Abrasive Waterjet Pressure (Awjp), Stand-off Distance (Sd), and Abrasive Flow Rate (Afr) on two critical responses: Surface Roughness (Sr) and Kerf Taper Angle (Kta). The optimum parameter setting (Awjp = 220 MPa, Sd = 2 mm, Afr = 230 g⋅min-1) minimized Sr and Kta. Experimental validation demonstrated that the ANN model obtained greater prediction accuracy with an average error of 1.2154%, compared to 12.18114% for the GRA model. Regression analysis produced R² = 95.05% and R²(adj) = 80.19%. The study demonstrates the effectiveness of ANN in optimizing AWJD processes and enhancing the machining performance of AZ31B-Mg alloy, that supports greater adoption of the alloys in high-performance engineering applications.

ABSTRACTGray jujube (Ziziphus jujuba Mill) is an important economic fruit crop in Xinjiang, China, whose fruit quality is regulated by complex interactions among tree architecture, physiological functions, and environmental factors. Based on 2 years of field experiments, we developed an interpretable artificial neural network model integrating 13 structural and physiological indicators to predict four quality parameters: vitamin C (VC), soluble sugar, titratable acid, and sugar‐acid ratio. The model architecture was optimized through Bayesian optimization, resulting in a 13–4–1/13–5–1 network structure with high prediction accuracy (R2 = 0.89–0.98). Biological interpretation of the connection weights revealed that the elongation of bearing shoots (1.2–3.1 cm/month) and SPAD values (33–41.5) were key drivers of VC accumulation, reflecting their roles in photosynthate transport and light‐harvesting efficiency. Canopy structural characteristics, particularly leaf inclination angles of 26°–34° combined with a direct beam transmittance of 0.32–0.43, were found to synergistically enhance sugar accumulation by optimizing light distribution while maintaining sufficient gas exchange. Furthermore, net photosynthetic rates exceeding 12 μmol·m−2·s−1 significantly reduced organic acid content, indicating a shift in carbon partitioning toward sugar synthesis. These findings demonstrate that the model successfully bridges computational analysis with biological processes, providing both a predictive tool and mechanistic insights for gray jujube quality management. The integration of architectural, physiological, and environmental parameters in this framework offers a comprehensive approach for precision cultivation of this important crop.

This study addresses the urgent need for an AI model to predict Anaplastic Lymphoma Kinase (ALK) inhibitors for Non-Small Cell Lung Cancer treatment, targeting the ALK-positive mutation. With only five Food and Drug Administration approved ALK inhibitors currently available, effective drugs remain in demand. Leveraging machine learning (ML) and deep learning (DL), our research accelerates the precise screening of novel ALK inhibitors using both ligand-based and structure-based approaches. In ligand-based approach, an ensemble voting model comprising three base learners to classify potential ALK inhibitors, achieving promising retrospective validation results. Notably, the ML-based XGBoost algorithm exhibited compelling results with external validation (EV)-f1 score of 0.921, EV-Average Precision (AP) of 0.961, cross-validation (CV)-f1 score of [Formula: see text] and CV-AP of [Formula: see text]. Besides, the DL-based Artificial Neural Network (ANN) model demonstrated comparative performance with EV-f1 score of 0.930, EV-AP of 0.955, CV-f1 score of [Formula: see text] and CV-AP of [Formula: see text]. For structure-based approach, an XGBoost consensus docking model utilized scores from three molecular docking programs (GNINA 1.0, Vina-GPU 2.0, and AutoDock-GPU) as features. Combining these two approaches, we virtually screened 120,571 compounds, identifying three promising ALK inhibitors, CHEMBL1689515, CHEMBL2380351, and CHEMBL102714, that bind to the protein's pocket and establish hydrophobic contacts in the hinge region through their ketone groups, resembling Alectinib's interaction. Comparative analysis revealed traditional ML models outperformed Graph Neural Networks (GNN), highlighting the critical role of feature engineering and dataset size importance. The study recommends further in vitro testing to validate the prospective screening performance of these models. A graphical user interface is available at https://huggingface.co/spaces/thechuongtrinh/ALK_inhibitors_classification .

Rising water demand has intensified pollution and created an urgent need for efficient treatment methods. Adsorption is a green and low-cost approach, yet conventional adsorbents often face sustainability and regeneration challenges. In this study, a novel adsorbent was developed by pyrolyzing Suaeda monoica leaf powder (LP) into biochar (BC300), followed by base treatment and coprecipitation with NiCuZnFe2O4 spinel to form a ferrite–biochar composite (FCOB). FCOB effectively removed Crystal Violet (CV) dye from aqueous solution. FE-SEM images revealed a layered morphology, while FTIR analysis confirmed multiple adsorption mechanisms, including hydrogen bonding, electrostatic attraction, surface complexation, and pore filling for CV adsorption. Optimization studies showed maximum CV removal at pH 8 with a 30 mg FCOB dose, maintaining >95% removal up to 200 mg L−1 dye concentration. For higher concentrations, 150, 250 mg L−1, the equilibrium time was 120 min. The Langmuir model indicated monolayer adsorption with a maximum capacity (qmax) of 325.5 mg g−1 at 30 °C, whereas the Dubinin–Radushkevich (D–R) model (E < 8 kJ mol−1) suggested physical adsorption. Kinetic analysis revealed that the pseudo-second-order (PSO) model best described the process, indicating the chemical nature of CV adsorption onto FCOB, while the Elovich model provided a better fit at higher concentrations, reflecting surface heterogeneity. Thermodynamic parameters confirmed that CV adsorption was spontaneous and endothermic (ΔH° = 49.03 kJ mol−1). FCOB retained >275 mg g−1 capacity after five regeneration cycles, demonstrating good reusability. Artificial neural network (ANN) modeling reliably predicted adsorption performance (R2 > 0.99) using pH, dye concentration, dose, time, and temperature as inputs. These findings highlight FCOB as an economical, eco-friendly, and scalable adsorbent for dye removal from wastewater.

This study proposes an efficient and accurate computational framework for the buckling analysis of Euler–Bernoulli microbeams modeled using Strain Gradient Elasticity (SGE) theory. The sixth-order governing equations derived from SGE are solved via a combination of the Initial Value Method and an Approximate Transfer Matrix formulation based on a truncated matricant series. This semi-analytical approach eliminates symbolic computation and enables precise evaluation of dimensionless critical buckling loads under various classical boundary conditions. A feedforward Artificial Neural Network (ANN) is trained on computed buckling data corresponding to γ / L values ranging from 1/10 to 1/20 and tested on intermediate values not included in the training. The ANN model demonstrates highly accurate predictions for unseen γ / L values, exhibiting strong agreement with the analytical buckling results. This combined framework not only overcomes the symbolic complexity associated with high-order differential equations but also provides an efficient and generalizable predictive model for microscale structural stability analysis. The results reveal that increasing the gradient coefficient significantly raises the critical buckling load, reflecting the size-dependent stiffening behavior inherent in SGE theory. The proposed methodology effectively integrates analytical rigor with data-driven efficiency, representing a significant and novel contribution to microscale structural mechanics.

This study develops a robust and efficient rainfall prediction model using an Artificial Neural Network (ANN), significantly enhanced through integrated data augmentation, regularization, and Bayesian optimization techniques. We utilized a dataset of 118 monthly rainfall records from Makassar City, spanning 2014–2022, sourced from the Meteorological, Climatological, and Geophysical Agency (BMKG). To effectively capture inherent temporal patterns, lag features (specifically lag-1, lag-3, and lag-6 rainfall values) were meticulously constructed as input variables. Subsequently, Min-Max normalization was applied across all features, ensuring input consistency and optimizing the ANN's learning process. An initial manual grid search identified the most effective baseline ANN architecture, featuring four hidden layers ([128, 32, 16, 64] neurons), a tanh activation function, and a learning rate of 0.01. While the baseline ANN model achieved a commendable initial performance with an RMSE of 0.1608, comprehensive experiments revealed the superior benefits of a fully integrated approach. This advanced model, which synergistically combined data augmentation (to address data limitations and enhance generalization), regularization (to mitigate overfitting), and Bayesian optimization (for efficient hyperparameter tuning), demonstrated significantly improved generalization capabilities and enhanced model stability. This integrated model yielded an RMSE of 0.1861, an MSE of 0.0346, and an MAE of 0.1359. These compelling findings unequivocally underscore that integrated optimization strategies are crucial for developing more robust and reliable ANN-based rainfall prediction models, particularly for critical applications in climate-based time series forecasting.

ABSTRACTModeling and estimating radon concentration are of crucial interest to support health protection campaigns. In the literature, many studies concentrated on indoor radon, while few of them investigated the outdoor radon spatial distribution and the factors that influence its formation. In this context, the vast possibilities of the artificial intelligence systems, based on machine learning techniques, can show remarkable capabilities. This paper focuses on the optimization of the architecture and the parameters of an artificial neural network (ANN) for inferring outdoor radon concentrations. More specifically, in the development of alternative ANN models, the Feed‐Forward Back propagation with the Levenberg–Marquardt is performed with different hidden layers to train the models and a bootstrap resampling method is applied to improve the model generalization. Some evaluation metrics and a sensitivity analysis are also included in order to assess the prediction accuracy among the ANN models.

This study investigates the steady, laminar motion of a non-Newtonian Oldroyd-B nanofluid over an inclined plate, integrating Buongiorno’s nanofluid model to account for Brownian motion and thermophoresis. The novel integration of couple stress and Forchheimer inertia in the analysis, coupled with advanced Bayesian-regularized ANN modelling, distinguishes this work. Governing equations are transformed using similarity variables and solved numerically via MATLAB’s bvp4c solver. The effects of couple stress, relaxation time, Forchheimer number, thermal radiation, thermophoresis, Brownian motion, and activation energy on velocity, temperature, and concentration profiles are systematically analyzed. Results reveal that couple stress and relaxation time reduce velocity, while thermal radiation and thermophoresis elevate temperature. Brownian motion decreases concentration, and activation energy influences both temperature and concentration oppositely. Multiple linear regression models quantify relationships between friction factor, Nusselt, and Sherwood numbers and key parameters, while a Bayesian-regularized artificial neural network (ANN) demonstrates high predictive accuracy (R-values ~1). It is noticed that increasing the couple stress parameter from 0.1 to 2.5 reduces friction factor by 59.8%, increasing the thermophoresis parameter from 0.1 to 2.5 decreases the Nusselt number by 7.8%, reflecting reduced heat transfer, and increasing the Brownian motion parameter from 0.1 to 2.5 reduces the mass transmission rate by 2.6%.

Solar thermal collectors represent a critical technology for sustainable energy harvesting, yet their optimal design remains challenging due to complex multi-physics interactions and numerous design parameters. This research presents a comprehensive computational framework that integrates Computational Fluid Dynamics (CFD) with data-driven optimization techniques to enhance solar thermal collector performance. The methodology combines advanced numerical modeling with machine learning algorithms to achieve both high prediction accuracy and computational efficiency. A validated three-dimensional CFD model was developed using ANSYS Fluent to simulate heat transfer phenomena within various collector geometries, including air-based systems with transverse triangular blocks and parabolic trough concentrators. The study generated 935 numerical cases across diverse operational parameters, which were subsequently used to train artificial neural networks (ANN), support vector regression (SVR), and linear regression models. The optimized ANN model achieved coefficient of determination values of 0.94, demonstrating superior predictive capabilities compared to traditional approaches. Entropy analysis identified thermal conductivity as the most influential parameter, contributing approximately 20% to overall thermal efficiency. The integrated approach successfully reduced computational time from over 1500 seconds for full CFD simulations to approximately 10 milliseconds for ANN predictions, while maintaining prediction accuracy within 6% of experimental data. Results indicate that collector designs incorporating heat transfer enhancement features such as triangular blocks and optimized geometric parameters achieve thermal efficiencies up to 68%, representing a 15% improvement over conventional configurations. Temperature distribution analysis revealed optimal operating ranges between 298K and 340K for maximum heat transfer effectiveness. The framework demonstrates significant potential for accelerating solar thermal system development while reducing computational costs and improving design optimization capabilities.

A neural model was developed to predict the distribution of ZnO nanoparticles obtained by electrochemical synthesis. It is a three-layer multilayer perceptron (MLP) artificial neural network (ANN) with five neurons in the input layer, eight neurons in the hidden layer, and one neuron in the output layer. This network has a hyperbolic tangent activation function for the neurons in the hidden layer and an exponential activation function for the neuron in the output layer. The input (independent) variables are particle size (nm), solvent type, and temperature (°C), and the output (dependent) variable is fraction share (%). The best neural model (ann08) has a root mean square error (RMSE) 0.84% for the training subset, 0.98% for the testing subset, and 1.27% for the validation subset. The RMSE values are therefore small, which enables practical use of the ANN model.

Purpose The present study investigated the effects of different input parameters such as cooling/lubrication conditions, cutting speed and feed rate on power consumption, surface roughness, cutting temperature, vibration and wear mechanisms during the milling of AISI 316 Ti alloy. Furthermore, the study aims to determine the estimation method that gives optimum results by estimating the output parameters using response surface methodology (RSM), artificial neural network (ANN) and adaptive network-based fuzzy inference system (ANFIS) methods. Besides these, it is aimed to obtain efficient results for the industry in the processing of AISI 316 Ti stainless steel. In addition, it is aimed to save time and cost for the manufacturing industry by determining the optimum cutting parameters and output parameters using modeling methods such as RSM, ANN and ANFIS. Design/methodology/approach AISI 316 Ti alloy was obtained commercially. Power consumption during machining under different cooling/lubrication conditions, cutting speeds, and feed rates was performed using a power analyzer connected to the machine’s electrical input. The average surface roughness values were determined by calculating the average of the measurements taken from different points perpendicular to the machining direction. An infrared camera was used to measure the temperature in the cutting zone during machining tests. Vibrations occurring during machining were measured using a digital vibration measurement system. RSM, ANN and ANFIS models were used to estimate the output parameters. Findings The minimum quantity lubrication (MQL) machining condition resulted in reduced power consumption, surface roughness and vibration values to decrease, while air machining condition led to a decrease in cutting temperature. It was observed that low cutting speed and feed rate caused low power consumption, cutting temperature and vibration. It was observed that increasing the cutting speed and decreasing the feed rate had a positive effect on the surface roughness. The study revealed that the reliability coefficient and predictive capability of the ANN model were higher than those of the RSM and ANFIS models. Social implications The study emphasizes the importance of predicting power consumption, surface roughness, cutting temperature and vibration during the machining of AISI 316 Ti alloy with a specific set of machine parameters, thereby reducing the cost and time associated with experimental trials. This advancement is expected to enhance the efficiency, cost-effectiveness, and product quality for manufacturing companies working with AISI 316 Ti alloy. Moreover, the findings reveal that the Artificial Neural Network (ANN) model demonstrates superior accuracy among the prediction models evaluated, indicating its potential for practical application in industrial settings. These improvements contribute not only to economic benefits but also support sustainable manufacturing practices by minimizing resource waste. Originality/value This study focuses on the effects of different cooling/lubrication conditions and cutting parameters on power consumption, surface roughness, vibration and wear mechanisms during the milling of AISI 316 Ti alloy. It is also a comprehensive investigation in which optimal machining parameters are determined by estimating output parameters using RSM, ANN and ANFIS models.

Abstract Stress concentration factors (SCFs) play a critical role in the structural integrity of circular plates subjected to internal and external compressive loading. This study uses Artificial Neural Networks (ANNs) and the Finite Element Method (FEM) to examine SCFs of circular plates with different circular hole configurations. The stress distribution and stacking patterns were thoroughly examined using FEM. Then, using geometrical and loading parameters, an ANN model was created that can predict SCFs. The FEM-generated data sets were used to train and validate the ANN model, which showed that it could generalize to various hole configurations and pressure levels. A comparison of the FEM and ANN findings revealed a strong connection, confirming the suggested ANN model’s dependability. The study’s conclusions offer valuable information for perforated plate design and optimization, offering a computationally effective substitute for traditional numerical methods.

Local scour around bridge piers is one of the primary causes of structural failure in bridges. Therefore, this study focuses on addressing the estimation of maximum scour depth (dsm), which is essential for safe and resilient bridge design. Many studies in the last eight decades have included metadata collection and developed around 80 empirical formulas using various scour-affecting parameters of different ranges. To date, a total of 33 formulas have been comparatively analyzed and ranked based on their predictive accuracy. In this study, novel formulas using semi-empirical methods and gene expression programming (GEP) have been developed alongside an artificial neural network (ANN) model to accurately estimate dsm using 768 observed data points collected from published work, along with eight newly conducted experimental data points in the laboratory. These new formulas/models are systematically compared with 74 empirical literature formulas for their predictive capability. The influential parameters for predicting dsm are flow intensity, flow shallowness, sediment gradation, sediment coarseness, time, constriction ratio, and Froude number. Performances of the formulas are compared using different statistical metrics such as the coefficient of determination, Nash–Sutcliffe efficiency, mean bias error, and root-mean-squared error. The Gauss–Newton method is employed to solve the nonlinear least-squares problem to develop the semi-empirical formula that outperforms the literature formulas, except the formula from GEP, in terms of statistical performance metrics. However, the feed-forward ANN model outperformed the semi-empirical model during testing and validation phases, respectively, with higher CD (0.790 vs. 0.756), NSE (0.783 vs. 0.750), lower RMSE (0.289 vs. 0.301), and greater prediction accuracy (64.655% vs. 61.935%), providing approximately 15–18% greater accuracy with minimal errors and narrower uncertainty bands. Using user-friendly tools and a strong semi-empirical model, which requires no coding skills, can assist designers and engineers in making accurate predictions in practical bridge design and safety planning.

Sports turfs and urban landscapes generate waste biomass in the form of grass clippings. Decomposing grass clippings can recycle nutrients to soil. However, decomposing can have adverse environmental effects such as gaseous emissions. The magnitude of air pollution caused by gaseous emissions from grass clippings is unknown. This research investigated CO, CO2, and O2 exchange during the decomposition of grass clippings. Emissions from grass clippings collected at four football fields with different levels of fertilization and agrotechnical treatments were studied. The mowed grass was collected throughout the spring-to-autumn football season. The results showed that grass clippings from sports turfs can generate up to 5 times more CO emissions compared to a mixture of grass and cattle manure. CO2 production and O2 consumption were relatively similar for all seasons, except for clippings from the unfertilized pitch. Artificial neural network (ANN) models predicted the CO and CO2 emissions resulting from the disposal of grass clippings with R2 for CO &amp;gt; 0.81 and CO2 &amp;gt; 0.98, respectively. This research contributes to emission inventories and highlights the relatively minor contribution from decomposing biomass.

Designing efficient nanoparticle-enhanced CO2 capture systems is challenging due to the diversity of nanoparticles, solvent formulations, reactor configurations, and operating conditions. This study presents the first ANN-based meta-analysis framework developed to predict CO2 absorption enhancement across multiple reactor systems, including batch reactors, packed columns, and membrane contactors. A curated dataset of 312 experimental data points was compiled from literature, and an artificial neural network (ANN) model was trained using six input variables: nanoparticle type, concentration, system configuration, base fluid, pressure, and temperature. The proposed model achieved high predictive accuracy (R2 &gt; 0.92; RMSE: 4.2%; MAE: 3.1%) and successfully captured complex nonlinear interactions. Feature importance analysis revealed nanoparticle concentration (28.3%) and system configuration (22.1%) as the most influential factors, with functionalized nanoparticles such as Fe3O4@SiO2-NH2 showing superior performance. The model further predicted up to 130% enhancement for ZnO in optimized membrane contactors. This AI-driven tool provides quantitative insights and a scalable decision-support framework for designing advanced nanoparticle–solvent systems, reducing experimental workload, and accelerating the development of sustainable CO2 capture technologies.

Artificial neural network modelling of reactive red 45 Azo dye removal by peroxi-electrocoagulation

The road construction sector is integral to economic growth in developing countries. However, road construction projects in Pakistan frequently experience budget overruns and delays, largely due to inaccurate cost prediction practices. These traditional techniques rely on expert judgement and analogies to past projects, which are often imprecise and subject to human biases. This study aims to enhance cost prediction process for road construction projects in Pakistan by developing a machine learning-based prediction model. To this end, a dataset of road construction projects is prepared using information from Planning Commission Forms I, obtained from several departments of the government of Pakistan. The dataset consists of 86 samples and includes macro-level features such as project duration, road dimensions, and road type. An Artificial Neural Network model is trained using this dataset to predict project costs. The model showed promising results, achieving MAPE and R2 scores of 49.6% and 0.89, respectively, and making better predictions than human estimates on over-budget projects. Interviews conducted with project managers for qualitative validation of the dataset and model also backed this hypothesis. These findings highlight the feasibility and benefits of adopting machine learning models to reduce budget overruns and enhance planning efficiency in road construction projects.

An integrated framework combining Finite Element Analysis (FEA) and Artificial Neural Networks (ANN) is presented to enhance the prediction and design of bioprinted scaffolds. By leveraging the strengths of data-driven learning and physics-based simulations, the hybrid approach (ANN + FEA) achieves superior predictive accuracy and generalization compared to standalone approaches. Validation against experimental results demonstrates that a single ANN model yields a relative error of 5.17% when predicting scaffold Young&amp;rsquo;s modulus. Incorporating FEA simulation based on ANN-predicted geometry and material properties reduces the relative error to 4.72%, representing an 8.6% improvement. The framework also enables the accurate simulation for unseen combinations of printing parameters located far from the experimental data manifold, reducing prediction errors from 14.2% (ANN-only) to 5.7% (hybrid). By integrating predictive modeling, simulation, and data augmentation, this approach offers an efficient pathway for optimizing scaffold designs and accelerating the development of biomaterials with tailored mechanical performance.

ABSTRACTA propensity weighted (PSW) methodology was recently proposed for assessing the treatment effect conditional to the probability of non‐specific response to a treatment (prob‐NSRT). Prob‐NSRT was estimated using an artificial neural network (ANN) model applied to pre‐randomization and study endpoint observations in a placebo arm of a placebo‐controlled clinical trial. Placebo data were initially used to estimate prob‐NSRT, then the ANN model was applied to the data of each individual in each treatment arm (placebo + active) for estimating the individual prob‐NSRT, and finally all data in the trial enriched by the prob‐NSRT values were used to assess the treatment effect. One of the major limitations of this methodology was that the ANN model was developed and applied to analyze data in the same dataset. To overcome this limitation, a new artificial intelligence driven nonlinear mixed effect modeling approach (AI‐NLME) is proposed. This approach involves the development of the ANN model using a dataset that is independent from the dataset used to estimate the treatment effect. A case study is presented using data from a randomized, placebo‐controlled trial in major depressive disorders. The AI‐NLME approach provided an effective tool for controlling the confounding effect of treatment non‐specific response, for increasing signal detection, for decreasing heterogeneity in the response, for increasing the effect size, for better assessing the responder rate, and for providing a reliable estimate of the “true” treatment effect. These findings provide convergent evidence on the potential role of AI‐NLME to become the reference approach for analyzing placebo‐controlled clinical trials.

Predicting diameter and tensile strength of electrospun fibers for biomedicine: A comparison of Box-Behnken design, traditional machine learning and deep learning.