Generalized Regression Neural Network Model Research Articles

A Comprehensive GRNN Model for the Prediction of Cutting Force, Surface Roughness and Tool Wear During Turning of CP-Ti Grade 2

Titanium alloys are most widely used in aerospace, military, marine, medical and bio-medical industries because of their unique properties such as high corrosion resistance, biocompatibility, superior strength-to-weight ratio and ability to withstand at elevated temperatures. In spite of the above mentioned capabilities these alloys are classified as “Difficult-to-machine” type materials. This is due to lower thermal conductivity and high chemical affinity of these materials. Poor thermal conductivity and increased chemical reactivity during titanium machining causes rapid tool wear. This might be contributed to the high heat generated at primary deformation zone. This situation necessitates proper selection of the machining parameter as well as cutting tool material before machining the workpiece. The present paper suggests a novel approximation tool for predicting some of the vital machinability aspects during the turning of commercially pure titanium grade 2. Experiments were conducted at five different cutting speeds (30, 60, 90, 120 and 150 m/min) with a constant feed rate of 0.12 mm/rev and depth of cut of 0.5 mm. The variations attained in the performance measures viz. Surface roughness (Ra), cutting force (Fc) and flank wear (VBc) were graphically analyzed. A generalized regression neural network (GRNN) model was developed for the prediction of aforesaid output responses. The neural network was trained, validated and tested with the experimentally measured data sets. The grid search method was employed to find the optimal sigma – value (kernel bandwidth) with minimum cross-validation error. The obtained results were validated and tested in order to demonstrate the adequacy of the model. The results indicated that the GRNN model is successfully capable of approximating turning quality characteristics within ± 5% error.

An optimized artificial neural network model for the prediction of rate of hazardous chemical and healthcare waste generation at the national level

This paper presents a development of general regression neural network (a form of artificial neural network) models for the prediction of annual quantities of hazardous chemical and healthcare waste at the national level. Hazardous waste is being generated from many different sources and therefore it is not possible to conduct accurate predictions of the total amount of hazardous waste using traditional methodologies. Since they represent about 40% of the total hazardous waste in the European Union, chemical and healthcare waste were specifically selected for this research. Broadly available social, economic, industrial and sustainability indicators were used as input variables and the optimal sets were selected using correlation analysis and sensitivity analysis. The obtained values of coefficients of determination for the final models were 0.999 for the prediction of chemical hazardous waste and 0.975 for the prediction of healthcare and biological hazardous waste. The predicting capabilities of the models for both types of waste are high, since there were no predictions with errors greater than 25%. Also, results of this research demonstrate that the human development index can replace gross domestic product and in this context even represent a better indicator of socio-economic conditions at the national level.

Investigation of thermal performance of unidirectional flow porous bed solar air heater using MLP, GRNN, and RBF models of ANN technique

In the present work, Multi-layer perceptron (MLP), Generalized regression neural network (GRNN), Radial basis function (RBF) and Multiple linear regression (MLR) models has been used to predict the thermal performance of unidirectional flow porous bed solar air heater. These four models have been constructed on the basis of actual experimental data and calculated values. Total 96 experimental data sets have been used in the present work. In GRNN, RBF and MLP models, six input parameters such as mass flow rate, wind speed, atmospheric temperature, inlet fluid temperature, fluid mean temperature and solar intensity were used in input layer, and one variable, the thermal efficiency was used in output layer. Same parameters were used in MLR model. It is observed that GRNN model is the best model due to lowest error and highest value of R2 as compared to MLP, RBF and MLR model performances. It is found that the value of MAE, RMSE and R2 for GRNN model are 1.1128E−03, 5.9284E−06 and 0.99758 respectively, and the model efficiency is 0.99760, which is the highest value as compared to other model. These results confirmed that the GRNN model is appropriate model to predict the thermal performance of solar air heater.

Prediction of the transition temperature of bent-core liquid crystals using fuzzy “digital thermometer” model based on artificial neural networks

A dataset containing transition temperature values for 243 bent-core liquid crystal (LC) compounds was used to develop quantitative structure–property relationship (QSPR) models using only 2D molecular descriptors and general regression neural network (GRNN). Beside a standard analogue GRNN model, another GRNN model with fuzzy digital response was created with the aim to estimate the prediction error for each compound. Two approaches for the selection of most relevant subset of descriptors, namely the partial mutual information (PMI) and self-organizing maps combined with chi square ranking, were also compared. The best results were obtained using analogue GRNN model based on PMI selected subset (R2=0.91), with the mean absolute error (MAE) lower in comparison with previously published corresponding QSPR models. The digital PMI-GRNN model enabled distinction between high and low accurate predictions, i.e. ones with absolute error higher than mean absolute error (MAE) and others with absolute error ≤ MAE, with the accuracy of 81%.

The influence of climatic inputs on stream-flow pattern forecasting: case study of Upper Senegal River

Ideal prediction and modeling of stream-flow and its hydrological applications are extremely significant for decision-making tasks and proper planning of water resource and hydraulic engineering. In the last two decades, the potential of soft computing approaches has increased dramatically in engineering and science problems. In this research, the utility of two soft computing approaches, namely support vector regression (SVR) model and generalized regression neural network (GRNN), is validated to predict 1 day ahead daily river flow data in the upper Senegal River basin at the Bafing Makana station in West Africa. The modeling is conducted by including the climatological information in the modeled stream-flow patterns. Correlation procedure is established and applied to obtain the modeling of the input variables with statistically significant lagged datasets at t − 1, t − 2, and t − 3 used as three input combination for each case study scenario. Different statistical indicators are used to evaluate the accuracy of the prediction models. The results show that the accuracy of the models varied by the scenario and the input datasets, where the SVR model yielded the best results for both modeling scenarios. It is also evident that combining the historical stream-flow data with the rainfall and evapotranspiration can ameliorate substantially the accuracy of the two models for predicting 1-day ahead stream-flow. A comparison of the optimal SVR and GRNN models in this problem indicates that SVR exhibits superior performance to the GRNN model in estimating the daily stream-flow data, irrespective of the modeling scenario and the datasets that is applied. The findings offer an opportunity to apply SVR model for predicting daily stream-flow, with less data requirement for the investigated Senegal River basin.

Predictive Analysis on Responses in WEDM of Titanium Grade 6 Using General Regression Neural Network (GRNN) and Multiple Regression Analysis (MRA)

In the present investigation two smart prediction tools, namely the general regression neural network (GRNN) and multiple regression analysis (MRA) models were developed to predict and compare some of the key machinability aspects like average kerf width, average surface roughness and material removal rate in the wire electrical discharge machining process of titanium grade 6. Pulse-on time, pulse-off time, wire feed and wire tension were considered as machining variables to develop the predictive model. In order to curtail cross-validation error in GRNN, optimized kernel bandwidth was utilized using the grid search method. The neural network and regression models were trained, validated and tested with measured data. A mathematical model was developed using multiple regression analysis. The ANOVA test was also conducted to determine the significant parameters affecting the responses. The results indicated that the predicted responses lie within ± 5% and ± 10% error for GRNN and MRA, respectively, which suggests that the GRNN model is more reliable and adequate than the regression model. A comparative study with previous research work was also done to confirm the novelty along with application potential of the proposed model.

Application of general regression neural network to model a novel integrated fluidized bed gasifier

In order to better understand gasification performance, a general regression neural network (GRNN) was developed to model a novel integrated fluidized bed (IFB) gasifier to research the correlative relationship between the input and output parameters of the IFB gasifier. Additionally, the prediction accuracy of the GRNN model was compared with the multivariate nonlinear regression (MNR) method. The performances of the two methods were evaluated using the mean relative error (MRE), the root mean square error (RMSE) and the coefficient of determination (R2). The GRNN model simulated the IFB gasifier with a higher R2, a lower RMSE and a lower MRE demonstrating the prediction accuracy of the GRNN model over the MNR method. Furthermore, the effects of the oxygen to coal ratio, the steam to coal ratio, the oxygen to fly ash ratio and the steam to fly ash ratio on gasification performance were analyzed using the proposed GRNN model.

Do the time series statistical properties influence the goodness of fit of GRNN models? Study on financial series

Recent researches on Generalized Regression Neural Networks show that this technique could be a promising option for modeling nonlinear time series, in general, and for financial series, in particular. Different types of artificial neural networks have been extensively studied, but the relationship between the statistical properties of the input data series and the models' accuracy was not emphasized. Therefore, our aim is to provide such an analysis. We study the Bucharest Exchange Trading series registered during the period from October 2000 to September 2014. Firstly, we test the series randomness, the existence of an increasing or nonlinear trend, its stationarity around a deterministic trend, and the breakpoints existence. Then, using the series decomposition, we define the detrended series and the deseasonalized series. Secondly, we build Generalized Regression Neural Network models for the original series, the subseries detected after the segmentation, the detrended, and deseasonalized ones. Comparing the modeling results, we conclude that some “regularity” properties (as normality and homoskedasticity) do not influence the models' quality (as expected).

Temperature-based modeling of reference evapotranspiration using several artificial intelligence models: application of different modeling scenarios

The establishment of an accurate computational model for predicting reference evapotranspiration (ET0) process is highly essential for several agricultural and hydrological applications, especially for the rural water resource systems, water use allocations, utilization and demand assessments, and the management of irrigation systems. In this research, six artificial intelligence (AI) models were investigated for modeling ET0 using a small number of climatic data generated from the minimum and maximum temperatures of the air and extraterrestrial radiation. The investigated models were multilayer perceptron (MLP), generalized regression neural networks (GRNN), radial basis neural networks (RBNN), integrated adaptive neuro-fuzzy inference systems with grid partitioning and subtractive clustering (ANFIS-GP and ANFIS-SC), and gene expression programming (GEP). The implemented monthly time scale data set was collected at the Antalya and Isparta stations which are located in the Mediterranean Region of Turkey. The Hargreaves–Samani (HS) equation and its calibrated version (CHS) were used to perform a verification analysis of the established AI models. The accuracy of validation was focused on multiple quantitative metrics, including root mean squared error (RMSE), mean absolute error (MAE), correlation coefficient (R2), coefficient of residual mass (CRM), and Nash–Sutcliffe efficiency coefficient (NS). The results of the conducted models were highly practical and reliable for the investigated case studies. At the Antalya station, the performance of the GEP and GRNN models was better than the other investigated models, while the performance of the RBNN and ANFIS-SC models was best compared to the other models at the Isparta station. Except for the MLP model, all the other investigated models presented a better performance accuracy compared to the HS and CHS empirical models when applied in a cross-station scenario. A cross-station scenario examination implies the prediction of the ET0 of any station using the input data of the nearby station. The performance of the CHS models in the modeling the ET0 was better in all the cases when compared to that of the original HS.

Modeling Relationship between Truck Fuel Consumption and Driving Behavior Using Data from Internet of Vehicles

AbstractIn this research, by taking advantage of dynamic fuel consumption–speed data from Internet of Vehicles, we develop two novel computational approaches to more accurately estimate truck fuel consumption. The first approach is on the basis of a novel index, named energy consumption index, which is to explicitly reflect the dynamic relationship between truck fuel consumption and truck drivers’ driving behaviors obtained from Internet of Vehicles. The second approach is based on a Generalized Regression Neural Network model to implicitly establish the same relationship. We further compare the two proposed models with three well‐recognized existing models: vehicle specific power (VSP) model, Virginia Tech microscopic (VT‐Micro) model, and Comprehensive Modal Emission Model (CMEM). According to our validations at both microscopic and macroscopic levels, the two proposed models have stronger performed in predicting fuel consumption in new routes. The models can be used to design more energy‐efficient driving behaviors in the soon‐to‐come era of connected and automated vehicles.

Nowcasting model of low wind profile based on neural network using SODAR data at Guarulhos Airport, Brazil

ABSTRACTA generalized regression neural network model was tested – as a nowcasting tool – to forecast the low wind profiles up to 45 min (i.e. at heights of 10, 100, 200, and 300 m) at the Guarulhos International Airport, São Paulo, Brazil. A data set representing over 4 years was generated from sonic detection and ranging and surface meteorological station, which registered vertical wind profiles with intervals from 10 m to approximately 500 m in height every 15 min, and surface meteorological variables were collected each minute, respectively. These data were simultaneously used to train, validate, and test the proposed model. The u and v forecasts generated at 300, 200, and 100 m were better than at 10 m, which could certainly be attributable to the surface roughness. In addition, the results also revealed that the performance of the model is time-dependent – decreasing over time – and that this may be correlated with the fact that the neural network is a statistical rather than physical model. The forecasts of wind components u and v are slightly biased (or closely matched to observations) at all heights, and forecast intervals with maximum values have median and average errors equal to 0.070 and −0.017 ms−1, respectively. The forecast model’s results were evaluated using the values of four categorical statistics: probability of detection; probability for non-events; bias; and false-alarm ratio, with respectable minimum and maximum values for u wind principal components equal to 0.841, 0.833, 0.159, 0.981 at 10 m for 45-min forecasts and 0.989, 0.987, 0.011, 0.999 at 300 m for 15-min forecasts, respectively.

Estimating Forest Carbon Fluxes Using Machine Learning Techniques Based on Eddy Covariance Measurements

Approximating the complex nonlinear relationships that dominate the exchange of carbon dioxide fluxes between the biosphere and atmosphere is fundamentally important for addressing the issue of climate change. The progress of machine learning techniques has offered a number of useful tools for the scientific community aiming to gain new insights into the temporal and spatial variation of different carbon fluxes in terrestrial ecosystems. In this study, adaptive neuro-fuzzy inference system (ANFIS) and generalized regression neural network (GRNN) models were developed to predict the daily carbon fluxes in three boreal forest ecosystems based on eddy covariance (EC) measurements. Moreover, a comparison was made between the modeled values derived from these models and those of traditional artificial neural network (ANN) and support vector machine (SVM) models. These models were also compared with multiple linear regression (MLR). Several statistical indicators, including coefficient of determination (R2), Nash-Sutcliffe efficiency (NSE), bias error (Bias) and root mean square error (RMSE) were utilized to evaluate the performance of the applied models. The results showed that the developed machine learning models were able to account for the most variance in the carbon fluxes at both daily and hourly time scales in the three stands and they consistently and substantially outperformed the MLR model for both daily and hourly carbon flux estimates. It was demonstrated that the ANFIS and ANN models provided similar estimates in the testing period with an approximate value of R2 = 0.93, NSE = 0.91, Bias = 0.11 g C m−2 day−1 and RMSE = 1.04 g C m−2 day−1 for daily gross primary productivity, 0.94, 0.82, 0.24 g C m−2 day−1 and 0.72 g C m−2 day−1 for daily ecosystem respiration, and 0.79, 0.75, 0.14 g C m−2 day−1 and 0.89 g C m−2 day−1 for daily net ecosystem exchange, and slightly outperformed the GRNN and SVM models. In practical terms, however, the newly developed models (ANFIS and GRNN) are more robust and flexible, and have less parameters needed for selection and optimization in comparison with traditional ANN and SVM models. Consequently, they can be used as valuable tools to estimate forest carbon fluxes and fill the missing carbon flux data during the long-term EC measurements.

Prediction and optimization of surface roughness and micro-hardness using grnn and MOORA-fuzzy-a MCDM approach for nitinol in WEDM

The unique ability of shape memory alloy or smart alloy to memorize its previous form has drawn notable attention recently in a wide range of commercial applications. In efforts to precisely predict and compare few significant WEDM machinability aspects like surface roughness [arithmetic mean roughness (Ra), root mean square roughness (Rq) and maximum peak-to-valley height (Rz)] and micro-hardness (MH) of shape memory alloy nitinol, general regression neural network (GRNN) model was developed. Five critical machining parameter, namely pulse-on time (TON), discharge current (I), wire feed (WF), wire tension (WT) and flushing pressure (FP) were taken as machining input for the experiments. The grid search method was employed to minimize cross-validation error. The developed GRNN model predicted the responses within ±10% error indicating GRNN as a competent strategy to predict WEDM responses. A multi-criteria decision making (MCDM) approach, Fuzzy logic coupled with multi-objective optimization on the basis of ratio analysis (MOORA) is introduced to optimize different correlated responses. The parametric combination, TON = 12 µs, I = 10 A, WT = 12 N, WS = 150 mm/s. and FP = 8 Bar, were found to yield the preferred results. ANOVA test shows the efficiency of this hybrid MCDM model. Confirmation test has been done to validate the optimum process combination which demonstrates the improvement in WEDM responses. FESEM micrographs identifies lump of debris, micro cracks, pockmarks and recast layer in the machined surfaces.

Reliability‐Based Fatigue Life Prediction for Complex Structure with Time‐Varying Surrogate Modeling

To improve the computational efficiency and accuracy of reliability‐based fatigue life prediction for complex structure, a time‐varying particle swarm optimization‐ (PSO‐) based general regression neural network (GRNN) surrogate model (called as TV/PSO‐GRNN) is developed. By integrating the proposed space‐filling Latin hypercube sampling technique and PSO‐GRNN regression function, the mathematical model of TV/PSO‐GRNN is studied. The reliability‐based fatigue life prediction framework is illustrated in respect of the TV/PSO‐GRNN surrogate model. Moreover, the reliability‐based fatigue life prediction of an aircraft turbine blisk under multiphysics interaction is performed to validate the TV/PSO‐GRNN model. We obtain the distributional characteristics, reliability degree, and sensitivity degree of fatigue failure cycle, which are useful for the turbine blisk design. By comparing the direct simulation (FE/FV model), RSM, GRNN, PSO‐GRNN, and TV/PSO‐GRNN, we observe that the TV/PSO‐GRNN surrogate model is promising to perform the reliability‐based fatigue life prediction of the turbine blisk and enhance the computational efficiency while ensuring an acceptable computational accuracy. The efforts of this study offer a useful insight for the reliability‐based design optimization of complex structure.

Pile‐Raft Settlements Prediction under Coupled Static‐Dynamic Loads Using Four Heuristic Regression Approaches

One of the main driving factors for structures’ evaluation is the foundation settlement. Measuring structures’ settlement in field is costly especially when heavy loads are applied. Settlement prediction models can be used to avoid the high cost of settlement field tests. Four advanced heuristic regression methods are developed and applied in this study to estimate raft foundations’ settlement, namely, multivariate adaptive regression splines (MARS), M5 model tree (M5Tree), generalized regression neural networks (GRNN), and support vector regression (SVR) techniques. Simulation of raft pile foundations is utilized to calculate the settlements of piles under the effect of static and dynamic loads. Previous studies are compared with the newly developed models. The results show that the four models can be used to accurately predict foundations’ settlements in the training stage. Also, the results reveal that the MARS and SVR models performed slightly better than the M5Tree and GRNN models in the testing stage and accordingly can be used to predict foundations’ settlement. The SVR model outperformed other models when few numbers of measurements are available.

Evaluation of the Impact of Data Uncertainty on the Prediction of Physiological Patient Deterioration

In this paper, we investigate the impact of data uncertainty on the prediction of a vital sign of abnormalities. We analyze the evolution of mean arterial blood pressure (MAP) and heart rate (HR) vital signs time series over a given forecasting window, whose values are classified according to a model of normality that models patient’s physiological deterioration. To this end, we developed a generic prediction framework that enables the deployment of different methods for data preparation, training, prediction, and classification so as to perform a set of experiments. To address our purpose, we selected a general regression neural network model (to predict vital sign trends) and a support vector machine (to recognize abnormalities) as the appropriate methods for a baseline analysis. However, other algorithm choices could be tailored in the defined prediction framework. We extract from the multiparameter intelligent monitoring in intensive care II waveform database a set of 302, 940 vital sign samples (organized as 459 records) to define training and testing data sets. From our experiments, we obtained evidence demonstrating the incidence of null values in the MAP/HR input time series on forecasting tasks. This was performed under different conditions in terms of null values in the observation window, and varying other factors like labeling methods and forecast window sizes. In our experiments, we concentrate on analyzing the extent of prediction performance under such conditions and identify that a tolerable level of null values is around 25% where it is possible to still derive forecast windows with an accuracy target $\geq 0.9$ for both MAP and HR vital signs. The presented evaluation contributes to acknowledge the importance of uncertainty conditions on the prediction performance of vital signs and paves the way for future extensions involving other prediction approaches.

Multiple-input–multiple-output general regression neural networks model for the simultaneous estimation of traffic-related air pollutant emissions

Traffic-related air pollutant emissions have become a global environmental problem, especially in urban areas. The estimation of pollutant emissions is based on complex models that require the use of detailed travel-activity data, which is often unavailable and in particular, in developing countries. In order to overcome this issue, an alternative multiple-input–multiple-output general regression neural network model, based on basic socioeconomic and transport related indicators, is proposed for the simultaneous prediction of sulphur oxides (SOx), nitrogen oxides (NOx), ammonia (NH3), non-methane volatile organic compounds (NMVOC) and particulate matter emissions at the national level. The best model, created using only six inputs, has MAPE (mean absolute percentage error) values on testing in the range of 12–15% for all studied pollutants, except NMVOC (MAPE = 21%). The obtained predictions for SOx, NH3 and PM10 emissions were in good agreement with the reported emissions (R2 ≥ 0.93), while the predictions for NOx and NMVOC are somewhat less accurate (R2 ≈ 0.85). It can be concluded that the presented ANN approach can offer a simple and relatively accurate alternative method for the estimation of traffic-related air pollutant emissions.

Predicting the cation exchange capacity of reservoir rocks from complex dielectric permittivity measurements

Dimensional analysis was performed to understand the physics of ionic dispersion in reservoir rocks and to identify the factors influencing the cation exchange capacity (CEC) of these rocks. Dimensional analysis revealed the existence of a general relation independent of the unit system between two dimensionless groups denoted as the cationic dispersion number [Formula: see text] and the conductivity number [Formula: see text]. The former group [Formula: see text] stands for the ratio of the CEC to the electrical double-layer dispersion. The latter group [Formula: see text] represents the ratio of the low-frequency ionic conductivity to the high-frequency electronic polarization. Complex dielectric permittivity measurements on 121 water-saturated sandstone and carbonate rock samples were used to validate the dimensionless groups. In retrospect, dimensional analysis was useful in identifying variables influencing the CEC of hydrocarbon rocks. In particular, these variables consist of rock porosity [Formula: see text], specific surface area, and five other parameters of the Cole-Cole function, which describes the frequency dependence of the complex permittivity of rock samples in the range 10–1300 MHz. The Cole-Cole function parameters are [Formula: see text], which is a characteristic relaxation time; [Formula: see text] is the so-called spread parameter; [Formula: see text] is the real DC conductivity of water-saturated rocks; and [Formula: see text] and [Formula: see text], which are the real numbers representing the static and the high-frequency dielectric permittivities of the water-saturated rock, respectively. A general regression neural network (GRNN) model was developed to predict the CEC of shaly sandstones and carbonate rocks as a function of the variables identified by the dimensional analysis as essential in predicting the CEC. The CEC prediction capability of the GRNN model has been tested with a blind data set, and it has been compared with the CEC prediction capability using a nonlinear regression model developed in this study and using a linear regression model available in the literature. The GRNN model outperformed both of these empirical models. With the GRNN model, it is possible to obtain reliable quantitative estimates of the CEC of shaly sandstone and carbonate rocks using nondestructive frequency-dependent dielectric permittivity measurements that are rapid, economic, and accurate. In return, accurate and fast estimates of the CEC are useful in many petroleum engineering applications. They can be used to identify clay types and can also be used to quantify the volume of hydrocarbon in shaly sands using well-log resistivity data. The results of this study represent a major advantage for formation evaluation, wellbore stability analysis, and designing stimulation jobs.

Comparison of two new intelligent wind speed forecasting approaches based on Wavelet Packet Decomposition, Complete Ensemble Empirical Mode Decomposition with Adaptive Noise and Artificial Neural Networks

The wind speed forecasting is an important technology for the management of the wind energy. In this study, a new hybrid framework using the WPD (Wavelet Packet Decomposition), the CEEMDAN (Complete Ensemble Empirical Mode Decomposition) and the ANN (Artificial Neural Network) is proposed for wind speed multi-step forecasting. In the proposed framework, the WPD is employed to decompose the original wind speed series into a series of sub-layers, while the CEEMDAN is adopted to further decompose all the obtained sub-layers into a number of IMFs (Intrinsic Mode Functions). Finally, three types of ANN models, including the BP (Back-propagation Neural Network) models, the RBF (Radial Basis Function Neural Network) models and the GRNN (General Regression Neural Network) models, are utilized to complete the predicting computation for the decomposed wind speed series, respectively. To investigate the prediction performance of the presented framework, nine models are included in the comparisons as: the BP model, the WPD-BP model, the WPD-CEEMDAN-BP model, the RBF model, the WPD-RBF model, the WPD-CEEMDAN-RBF model, the GRNN model, the WPD-GRNN model and the WPD-CEEMDAN-GRNN model. Two experimental results indicate that: the proposed WPD-CEEMDAN-ANN models have better performance than the involved corresponding ANN models and WPD-ANN models in three-step predictions.

A Comparative Assessment of Artificial Neural Network, Generalized Regression Neural Network, Least-Square Support Vector Regression, and K-Nearest Neighbor Regression for Monthly Streamflow Forecasting in Linear and Nonlinear Conditions

Monthly forecasting of streamflow is of particular importance in water resources management especially in the provision of rule curves for dams. In this paper, the performance of four data-driven models with different structures including Artificial Neural Network (ANN), Generalized Regression Neural Network (GRNN), Least Square-Support Vector Regression (LS-SVR), and K-Nearest Neighbor Regression (KNN) are evaluated in order to forecast monthly inflow to Karkheh dam, Iran, in linear and non-linear conditions while the optimized values of the model parameters are determined in the same condition via the Leave-One-Out Cross Validation (LOOCV) method. Results show that the performance of the models is different in linear and nonlinear conditions; the cumulative ranking of the models according to the three assessment criteria including NSE, RMSE and R2 indicates that ANN performs best in linear conditions while LS-SVR, GRNN and KNN are in the next ranks, respectively. But in nonlinear conditions, the best performance belongs to LS-SVR, followed by KNN, ANN, and GRNN models.