Group Method Of Data Handling Research Articles

Regression modeling and multi-objective optimization of rheological behavior of non-Newtonian hybrid antifreeze: Using different neural networks and evolutionary algorithms

The research used an artificial neural network (ANN) model to examine the rheological properties of hybrid non-Newtonian ferrofluids (HNFFs) composed of Fe-CuO, water, and ethylene glycol. The performance of neural network was optimized using seven regression methods (RMs), namely Group Method of Data Handling (GMDH), Decision Tree (D-Tree), Multi-Layer Perceptron (MLP), Support Vector Machine (SVM), Extreme Learning Machine (ELM), Radial Basis Function (RBF), and Multiple Linear Regression (MLR). The findings highlighted GMDH method's superior performance when compared to neural networks. R and RMSE values attained by GMDH for the objective function (OF) μnf were 0.99436 and 2.0135, respectively. For the torque function OF, the values were 0.97652 and 4.8952. Margin of difference (MOD) calculations across various algorithms, such as MLP, SVM, RBF, D-Tree, ELM, MLR, and GMDH-Algos revealed significant disparities, indicating GMDH's efficacy. Comparison of R, RMSD, and standard deviation values between GMDH and MLR algorithms further underscored performance discrepancies. Specific parameters for which NSGA II Algo was rated highest among evaluation indices were as follows: a crossover rate of 0.7, a mutation rate of 0.02, a population size of 50, and 500 generations. Post-optimization, optimal values for μnf and torque (To) were determined as 6.595 and 3.543, respectively, with corresponding values for φ, T, and γ obtained as 0.185, 49.372, and 3.163, respectively. This comprehensive analysis sheds light on the effectiveness of various regression methods in modeling the rheological behavior of hybrid non-Newtonian ferrofluids, contributing to advancements in fluid dynamics research.

Accurate neural-network-based fitting of full-dimensional N2 –Ar and N2–CH4 two-body potential energy surfaces aimed at spectral simulations

We describe the development of machine-learned (ML) potentials for flexible, weakly interacting monomers. A recently suggested permutationally invariant polynomial neural network (PIP-NN) approach is utilised to represent the full-dimensional two-body component of the molecular pair energy. To ensure the asymptotic zero-interaction limit, a tailored subset of the full invariant polynomial basis set is utilised, and their variables are modified to achieve a better fit of the correct asymptotic behaviour at a long range. This new technique is used to build full-dimensional potentials for the two-body N 2 –Ar and N 2 –CH 4 interactions by fitting databases of ab initio energies calculated at the coupled-cluster level of theory. The second virial coefficient, fully accounting for molecular flexibility, is then calculated within the classical framework using the obtained PIP-NN potential surfaces. A trajectory-based simulation of the N 2 –Ar collision-induced absorption is conducted, covering both the far- and mid-infrared ranges.

Machine learning prediction and characterization of sigma-free high-entropy alloys

Precipitation hardening provides a great potential to achieve superior strength-ductility trade-off in high-entropy alloys (HEAs). However, the formation of some precipitates such as brittle sigma phases may significantly deteriorate the ductility. Empirical criteria such as PSFE, VEC, and Md were proposed to predict the formation of undesirable phases during aging, nevertheless, they are insufficient to describe their stability and are not dependable for the noted purpose. Accordingly, for the first time, machine-learning models were conducted for this prediction. A feature selection based on Pearson correlation and mutual information has been used to improve the performance of models. Linear regression, second-order polynomial regression, decision tree, and neural network are conducted by machine learning methods in this paper. Among the implemented methods, the neural network had the best performance which improved the accuracy of the prediction by ∼20% compared to the conventional methods. Additionally, a new parameter based on linear regression, which is more reliable and user-friendly than thermodynamic parameters was developed in this investigation. The results showed that some elements such as Cr, Mo, and V may promote sigma precipitation. A HEA with a sigma-free microstructure was developed for validation of the model. We strongly believe that this HEA has a great potential for overcoming the strength-ductility trade-off because of not only the lack of sigma precipitate but also the presence of NiAl precipitate. The results are a step forward in designing alloys with the potential of microstructure engineering to achieve superior properties.

Optimal operation of the dam reservoir in real time based on generalized structure of group method of data handling and optimization technique

The historical data on water intake into the reservoir is collected and used within the framework of a deterministic optimization method to determine the best operating parameters for the dam. The principles that have been used to extract the best values of the flow release from the dam may no longer be accurate in the coming years when the inflow to dams will be changing, and the results will differ greatly from what was predicted. This represents this method’s main drawback. The objective of this study is to provide a framework that can be used to guarantee that the dam is running as efficiently as possible in real time. Because of the way this structure is created, if the dam’s inflows change in the future, the optimization process does not need to be repeated. In this case, deep learning techniques may be used to restore the ideal values of the dam’s outflow in the shortest amount of time. This is achieved by accounting for the environment’s changing conditions. The water evaluation and planning system simulator model and the MOPSO multi-objective algorithm are combined in this study to derive the reservoir’s optimal flow release parameters. The most effective flow discharge will be made feasible as a result. The generalized structure of the group method of data handling (GSGMDH), which is predicated on the results of the MOPSO algorithm, is then used to build a new model. This model determines the downstream needs and ideal release values from the reservoir in real time by accounting for specific reservoir water budget factors, such as inflows and storage changes in the reservoir. Next, a comparison is drawn between this model’s performance and other machine learning techniques, such as ORELM and SAELM, among others. The results indicate that, when compared to the ORELM and SAELM models, the GSGMDH model performs best in the test stage when the RMSE, NRMSE, NASH, and R evaluation indices are taken into account. These indices have values of 1.08, 0.088, 0.969, and 0.972, in that order. It is therefore offered as the best model for figuring out the largest dam rule curve pattern in real time. The structure developed in this study can quickly provide the best operating rules in accordance with the new inflows to the dam by using the GSGMDH model. This is done in a way that makes it possible to manage the system optimally in real time.

Artificial intelligence -based prediction of heat transfer enhancement in ferrofluid flow under a rotating magnetic field: Experimental study

The utilization of various techniques, both passive and active, to enhance heat transfer in fluids by scientists is continually advancing. One of the active methods being explored is the impact of a magnetic field (B) on flow to improve heat transfer. This study focuses on experimentally evaluating the effectiveness of B and the rotation of B on heat transfer of ferrofluid. The findings indicate that by increasing the volume fraction of nanoparticles, heat transfer can be enhanced by approximately 2.73–2.82 %. Furthermore, the use of B and intensifying its intensity leads to a 3.75–3.8 % enhancement in heat transfer. Rotating the B and increasing the rotational speed also contribute to improved heat transfer, with a 0.2 rad/s increase in rotation speed resulting in a 5%–12.43 % improvement in heat transfer. By utilizing available data and employing artificial intelligence methods such as Group Method of Data Handling (GMDH) and Long Short-Term Memory (LSTM), an estimation of the Nusselt number (Nu) has been achieved. The outcomes demonstrate that both models have displayed high accuracy in predicting Nu, with GMDH showing superior accuracy compared to LSTM in estimating Nu.

Evaluation and intelligent forecasting of energy and exergy efficiencies of a nanofluid-based filled-type U-pipe solar ETC using three machine learning approaches

The thermodynamic modeling of a filled layer U-type was conducted considering the six water-based and Engine Oil (EO)-based nanofluids (NFs). The results demonstrated that Fe2O3/EO NF is efficient from the exergy efficiency point of view. While MWCNT/water NF outperforms from the energy efficiency aspect. Moreover, the outcome indicated that when (Tin−Tamb)/It increased above 0.04 (K m2/W), Ltube has no considerable effect on the exergy efficiency. Still, the length impact is noticeable regarding outlet temperature and energy efficiency. Furthermore, the advantage of the filled-type over the conventional ETSC is more prominent at low m˙ (e.g. 4.0 kg/h). So that ηex in the filled type with Fe2O3/EO is improved by 2.29 % and 3.70 %, respectively, for (Tin−Tamb)/It of 0 and 0.04. Moreover, the increase in m˙ of the water/based and oil-based NFs, respectively, between 6 and 10 kg/h and 8–20 kg/h is suggested from the energy efficiency aspect. The results of machine learning modeling revealed that three models namely group method for data handling (GMDH), response surface method (RSM), and modified response surface model (MRSM) are capable of estimating ηen and ηex. While MRSM outperforms the two other models in terms of (R2 | ηen = 0.9986 and RMSE| ηen = 0.7849; R2 |ηex = 0.9988 and RMSE| ηex= 0.1513).

Modeling of ionic liquids viscosity via advanced white-box machine learning.

Ionic liquids (ILs) are more widely used within the industry than ever before, and accurate models of their physicochemical characteristics are becoming increasingly important during the process optimization. It is especially challenging to simulate the viscosity of ILs since there is no widely agreed explanation of how viscosity is determined in liquids. In this research, genetic programming (GP) and group method of data handling (GMDH) models were used as white-box machine learning approaches to predict the viscosity of pure ILs. These methods were developed based on a large open literature database of 2813 experimental viscosity values from 45 various ILs at different pressures (0.06-298.9MPa) and temperatures (253.15-573K). The models were developed based on five, six, and seven inputs, and it was found that all the models with seven inputs provided more accurate results, while the models with five and six inputs had acceptable accuracy and simpler formulas. Based on GMDH and GP proposed approaches, the suggested GMDHmodel with seven inputs gave the most exact results with an average absolute relative deviation (AARD) of 8.14% and a coefficient of determination (R2) of 0.98. The proposed techniques were compared with theoretical and empirical models available in the literature, and it was displayed that the GMDH model with seven inputs strongly outperforms the existing approaches. The leverage statistical analysis revealed that most of the experimental data were located within the applicability domains of both GMDH and GP models and were of high quality. Trend analysis also illustrated that the GMDH and GP models could follow the expected trends of viscosity with variations in pressure and temperature. In addition, the relevancy factor portrayed that the temperature had the greatest impact on the ILs viscosity. The findings of this study illustrated that the proposed models represented strong alternatives to time-consuming and costly experimental methods of ILs viscosity measurement.

Predicting Model Biases of the APOLLO3 Neutronics Code Using Machine Learning: Toward a Multiscale and Multifidelity Approach

As part of the verification, validation, and uncertainty quantification process applied to neutronics deterministic codes, there is a requirement to expand the validation domain, especially to accommodate new third-generation reactors. The objective of the present work is to estimate the numerical biases arising from the several approximations used in deterministic codes and across different points in the phase space. Typically, this is accomplished by comparing the deterministic code to be validated with a Monte Carlo or stochastic reference code (without significant approximations). Since these reference calculations are computationally expensive, this paper proposes an alternative approach for predicting model biases of the APOLLO3® deterministic code for third-generation pressurized water reactors using machine learning algorithms. Three types of metamodels are employed (polynomial regression, kriging, and neural networks). Two scales are investigated, from a single assembly to a cluster of 3 × 3 assemblies [small two-dimensional (2-D) core], with model biases evaluated for APOLLO3 schemes with various levels of accuracy (lattice and core solvers, with high- to low-fidelity approaches). For the small 2-D core, numerical biases are observed for reactivity and power peak, representing both global and local quantities of interest. Throughout the study, the best results are achieved using kriging or neural networks, even if polynomial regression provides satisfactory predictions in some cases. The possibility of predicting biases for different quantities is also introduced. In conclusion, this paper discusses the prospects of extending the applicability of these metamodels to small and large third-generation pressurized water reactor cores, the idea being to potentially use these metamodels to support safety demonstrations for the new reactors in the long term.

SHM Implementation on a RPV Airplane Model Based on Machine Learning for Impact Detection

AbstractIn this study, an on-working structural health monitoring system for impact detection on remote piloted vehicle (RPV) airplane is proposed. The approach is based on the propagation of Lamb waves in metallic structures on which Pb[ZrxTi1−x]O3 (PZT) sensors are bonded for receiving vibrational signals due to impact events. The proposed method can be used to detect impacts in aerospace structures, i.e. skin fuselage and/or wing panels. After the detection, machine learning (ML) algorithms (polynomial regression and neural networks) are applied for processing the acquired ultrasounds waves in order to characterise the impacts, in terms of time of flight (ToF) and relative location. Several test cases are studied: the ML models are tested both without external noise (in laboratory) and introducing external RC engine vibration (on-working conditions). Furthermore, this work presents the implementation of a mini-equipment for acquisition and data processing based on Raspberry Pi. A good agreement between laboratory and in-flight results is achieved, in terms of distance between the actual and calculated impact location.

Optimizing Activated Carbon Production from Waste Cashew Nut Shell with Zinc Chloride: A Box-Behnken Design and Group Method of Data Handling (GMDH) Application

In this study, Response surface methodology (RSM) and innovative Group Method of Data Handling (GMDH) approaches are applied to investigate the optimal process conditions of zinc chloride activated cashew nut production process. The effects of activation conditions (i.e. activation temperature, activation time, and impregnation ratio) on the achievable BET surface areas were studied with the aid of Box Behnken Design (BBD) and GMDH. Comparative analyses of RSM and GMDH-type neural models were further researched. During the process, the polynomial model equations developed were modified and fine-tuned to predict the highest BET surface area(s) using regression analysis and GMDH multi-layered iterative algorithm (MIA). Analysis of Variance (ANOVA) revealed that the significant factor(s) were impregnation ratio, impregnation ratio product, and the 2-way interactions (activation temperature and impregnation ratio) for ZnCl2 activated cashew nut shell. The best activation conditions for producing highest BET surface area of 504 m2.g-1 was activation temperature (873K), activation time (60 min), and impregnation ratio (1.50).The proposed GMDH-type BET model was ascertained to be the best model with average correlation coefficient (R) and root mean square error (RMSE) of 0.925 and 32.0 respectively. Sensitivity analysis conducted for GMDH-type neural network also revealed that the activation temperature and activation time with sensitivity values of 90.6% and 74.1% respectively were the most influential parameters in the basic (ZnCl2) activation process. The results of this study show that RSM and GMDH-type neural network could be applied as effective analytical tools for optimizing the ZCNS (zinc chloride-activated cashew nut shells) manufacturing process.

Estimation of soil organic carbon by combining hyperspectral and radar remote sensing to reduce coupling effects of soil surface moisture and roughness

Soil organic carbon (SOC) is important in the global carbon cycle. Accurate estimation of SOC content in cultivated land is a prerequisite for evaluating the carbon sequestration potential and quality of soils. However, existing SOC prediction studies based on hyperspectral remote sensing neglect the spectral response of the physical properties of surface soil, leading to inadequate model generalization. With the exponential growth of remote sensing data, the development of pixel-level soil spectral correction methods based on multi-source remote sensing data has become an interesting and challenging topic. This method aims to minimize the effect of soil physical properties on spectra, thus addressing the poor spatiotemporal transferability of SOC prediction models due to uncertain variations in surface soil physical properties. In this study, a soil spectral correction strategy is constructed using satellite hyperspectral image (HSI) and synthetic aperture radar (SAR) images through multi-order polynomial regression and convolutional neural networks. This strategy considers soil physical variables such as soil moisture (SM) content and root mean square height (RMSH) of soil surface roughness. The soil spectral correction model and SOC content prediction model were established using 80 soil samples collected from Site 1. Afterward, the performance and transferability of both models were verified using the remaining 25 samples from Site 1 and 50 samples from Site 2. The results showed that: 1) The effect of SM and RMSH on the soil pixel spectrum can be significantly reduced after correcting HSI using soil spectral correction strategy. The correlation coefficients between the corrected pixel spectrum and the ground-based spectrum increase by over 60 % compared with those between the original spectrum and the ground-based spectrum. 2) Soil spectral correction improves the prediction accuracy and mapping capability of HSI for SOC content, with the highest RP2 of 0.743 and RMSEP of 3.455 g/kg at Site 1. 3) Compared with the original HSI-based SOC prediction model, the soil spectral correction strategy based on multi-order polynomial and convolutional neural network reduced the RMSEP of SOC prediction results at Site 2 by 5.082 g/kg and 5.454 g/kg, and the RP2 increased by 0.390 and 0.409, respectively. 4) When predicting SOC content from raw HIS, SM and RMSH contribute to more than 60 % of the bias, with SM having a larger bias than RMSH. The findings of this study emphasize the influence of soil physical properties on SOC prediction and contribute to the existing research on SOC mapping using HSI and SAR data.

Predictive modelling of nitrogen dioxide using soft computing techniques in the Agra, Uttar Pradesh, India

Nitrogen dioxide (NO2) is one of the air pollutants which aggravates the human health as well as causes environmental issues. It is more causes respiratory problems due to acid rains. Agra is a major tourist destination spot in India also similarly air pollution also increased growing urbanization and traffic reflux. The current study aims to predicted the NO2 episodes in the Agra city using soft computing models namely, M5P, Random Forest (RF), Group method of data handling (GMDH), Multivariate adaptive regression (MARS), Reduced error pruning tree (REP Tree) and Random tree (RT). The models were generated using 1116 observations, from 2015 to 2020 with input parameters such as Particulate matter (PM2.5), Nitrogen monoxide (NO), Oxides of nitrogen (NOX), Sulphur dioxide (SO2), Carbon monoxide (CO), Ozone (O), Benzene (Be), Toluene, Relative humidity (RH), Wind speed (WS), Wind direction (WD), Solar radiation (SR), Barometric pressure (BP) and Xylene. The performance of each model was evaluated based on the six statistical indices, namely correlation coefficient (CC), root mean square error (RMSE), mean absolute error (MAE), normalized root of mean squared relative error (NRMSE), Willmott's Index (WI), and Legates and McCabe's Index (LMI). The performance evaluation models results of study area, Box plot and Taylor's diagram indicated that M5P is the outperforming model among others with testing CC = 0.9543, RMSE = 5.8006, MAE = 3,9204, NRMSE = 0.1512, WI = 0.9744, and LMI = 0.7549. Based on the sensitivity analysis indicated that NOx is the most influential parameter followed by WD and CO. These results of study area can be helpful to understanding the air pollution causes, health issues, and future NO2 levels around the study area with useful results for air pollution monitoring policy and development.

Application of group method of data handling and gene expression programming to modeling molecular diffusivity of CO2 in heavy crudes

Successful enhancement of heavy oil and bitumen production using CO2-based enhanced oil recovery (EOR) techniques requires extensive knowledge about the diffusivity of CO2 in heavy crudes. In this work, two advanced correlative approaches, namely group method of data handling (GMDH) and gene expression programming (GEP), were utilized to establish simple-to-use mathematical correlations by applying a comprehensive dataset including 260 laboratory data of CO2 diffusion coefficient (DC) in heavy crudes and taking into account all the parameters affecting the accuracy of the models. Moreover, the response surface and contour plots were analyzed to obtain the desired response values and operating conditions. The GMDH correlation represented more reliable results with better statistical criteria including average absolute percentage error (AAPRE) values of 7.47%, 7.83%, and 7.54% for training, testing, and the total sets, respectively. Also, the results showed that the increase in pressure and/or temperature caused an increment in the amount of diffusivity of CO2 in heavy crudes; and the maximum amount of CO2 DC was obtained at the upper limit of these operational parameters. At a given temperature and pressure, it appears that the highest CO2 DC in heavy crudes was obtained when the CO2 mass fraction was less than 0.02. Furthermore, the Pearson and Spearman correlation coefficients were used to investigate linear and non-linear relationships between input variables and the responses of GMDH and GEP correlations. The mass fraction of CO2, temperature, pressure, and temperature_PVT had a direct relationship, and on the other hand, the viscosity and density of crudes had a reverse relationship with CO2 DC. Finally, the reliability of the data bank of CO2 DC in heavy crudes along with the high validity of GMDH and GEP correlations was verified by the leverage technique. Notably, no out-of-leverage data were identified, with only seven and five data points considered as suspected for GMDH and GEP models, respectively.

TMPNN: High-Order Polynomial Regression Based on Taylor Map Factorization

The paper presents Taylor Map Polynomial Neural Network (TMPNN), a novel form of very high-order polynomial regression, in which the same coefficients for a lower-to-moderate-order polynomial regression are iteratively reapplied so as to achieve a higher-order model without the number of coefficients to be fit exploding in the usual curse-of-dimensionality way. This method naturally implements multi-target regression and can capture internal relationships between targets. We also introduce an approach for model interpretation in the form of systems of differential equations. By benchmarking on Feynman regression, UCI, Friedman-1, and real-life industrial datasets, we demonstrate that the proposed method performs comparably to the state-of-the-art regression methods and outperforms them on specific tasks.

Field Telemetry Drilling Dataset Modeling with Multivariable Regression, Group Method Data Handling, Artificial Neural Network, and the Proposed Group-Method-Data-Handling-Featured Artificial Neural Network

This paper presents data-driven modeling and a results analysis. Group method data handling (GMDH), multivariable regression (MVR), artificial neuron network (ANN), and new proposed GMDH-featured ANN machine learning algorithms were implemented to model a field telemetry equivalent mud circulating density (ECD) dataset based on surface and subsurface drilling parameters. Unlike the standard GMDH-ANN model, the proposed GMDH-featured ANN utilizes a fully connected network. Based on the considered eighteen experimental modeling designs, all the GMDH regression results showed higher R-squared and minimum mean-square error values than the multivariable regression results. In addition, out of the considered eight experimental designs, the GMDH-ANN model predicts about 37.5% of the experiments correctly, while both algorithms have shown similar results for the remaining experiments. However, further testing with diverse datasets is necessary for better evaluation.

Plant species recognition with optimized 3D polynomial neural networks and variably overlapping time–coherent sliding window

Plant species recognition with optimized 3D polynomial neural networks and variably overlapping time–coherent sliding window

An improved asynchronous batch gradient method for ridge polynomial neural network

The ridge polynomial neural network composed of pi-sigma modules is a typical higher-order feedforward network, which has good non-linear mapping capabilities. Due to the strong coupling of the network structure, the synchronous gradient method can easily result in significant fluctuations in updated weights. This will reduce the generalization ability of the ridge polynomial neural network for solving classification problems. In this paper, the proposed method is a novel improved asynchronous batch gradient method, and combined with the adaptive parameters of the activation function are trained synchronously. We strictly prove the convergence theorem of the improved method. The numerical experimental results also indicate that our method for training ridge polynomial neural networks can help the weight changes smoothly and the network has good generalization ability. The feasibility and effectiveness of our method are verified from both theoretical and experimental perspectives.

Hydrological dynamics of the Kalisindh and Parbati Rivers: An integrated analysis in the context of the Eastern Rajasthan Canal Project (ERCP)

This study conducts a comprehensive analysis of hydrological patterns in the Kalisindh and Parbati Rivers, highlighting the Eastern Rajasthan Canal Project (ERCP) as pivotal for bolstering regional water security. Employing an array of data sources, this research utilizes Polynomial Regression and neural network forecasting to dissect flow patterns, identifying significant virgin flow peaks in the mid-1980s and early 2000s for the Kalisindh River, and a notable peak in 2006–2007 for the Parbati River. Analysis reveals that the specific discharge rate of the Parbati River is diminishing at twice the rate of the Kalisindh River, with annual decreases of approximately 0.0038 cumecs/km2 for Parbati, compared to 0.0019 cumecs/km2 for Kalisindh. Furthermore, runoff volumes indicate that the Parbati River, specifically at the Khatoli Gauge & Discharge (G&D) site, experiences significantly higher runoff—28,137.912 million cubic meters (MCM)—in contrast to 15,795.094 MCM for the Kalisindh River at the Barod G&D site. The findings accentuate the necessity for science-based water management strategies to effectively combat water scarcity and climate change impacts. The ERCP emerges as a crucial initiative for sustainable management of the Kalisindh and Parbati Rivers in Rajasthan, underscoring its potential to serve as a model for sustainable water and river basin management.

New Full-Dimensional Reactive Potential Energy Surface for the H4 System.

As the most abundant molecule in the universe, collisions involving H2 have important implications in astrochemistry. Collisions between hydrogen molecules also represent a prototype for assessing various dynamic methods for understanding fundamental few-body processes. In this work, we develop a new and highly accurate full-dimensional potential energy surface (PES) covering all reactive channels of the H2 + H2 system, which extends our previously reported H2 + H2 nonreactive PES [J. Chem. Theory Comput., 2021, 17, 6747] by adding 39,538 additional ab initio points calculated at the MRCI/AV5Z level in the reactive channels. The global PES is represented with high fidelity (RMSE = 0.6 meV for a total of 79,000 points) by a permutation invariant polynomial neural network (PIP-NN) and is suitable for studying collision-induced dissociation, single-exchange, as well as four-center exchange reactions. Preliminary quasi-classical trajectory studies on the new PIP-NN PES reveal strong vibrational enhancement of all reaction channels.

Generalized structure of the group method of data handling for modeling iceberg drafts

The iceberg draft prediction is vital to mitigate the collision risk of deep keel icebergs with the seafloor-founded infrastructures, including the subsea pipelines, wellheads, hydrocarbon loading equipment, and communication cables crossing the Arctic and subarctic areas since the drifting icebergs may gouge the ocean floor and the physical and operational integrity of the submarine structures would be threatened. In this study, the iceberg drafts were simulated using the generalized structure of the group method of data handling (GS-GMDH) algorithm for the first time. The parameters affecting the iceberg drafts were determined, and five GS-GMDH models comprising GS-GMDH 1 to GS-GMDH 5 were developed utilizing those parameters governing. A dataset comprising 161 distinct case studies measured in the most significant field investigations of iceberg characteristics was generated, and the GS-GMDH models were trained through 60 % of the data, the rest of the data (i.e., 40 %) were considered for the GS-GMDH models’ validation. By defining different scenarios, the most accurate GS-GMDH model and the most important input parameters were identified. The sensitivity analysis demonstrated that the iceberg width ratio (W/H) and the iceberg shape factor (Sf) were identified as the most influencing input parameters. The comparison between the performance of the premium GS-GMDH model and the group method of data handling (GMDH), artificial neural network (ANN) algorithms, and the empirical models proved that the GS-GMDH model simulated the iceberg drafts with the highest level of precision and correlation along with the lowest degree of complexity. Based on the partial derivative sensitivity analysis (PDSA), the magnitude of iceberg drafts grew by increasing the value of the iceberg width and iceberg length. Ultimately, a GS-GMDH-based equation was presented to estimate the iceberg drafts for practical applications, particularly in the early stages of iceberg management projects and engineering designs.