Adaptive Neuro-fuzzy Inference System Network Research Articles

Predicting the AC conductivity of semiconductor composition thin films using ANFIS model; an integrated experimental and theoretical approach

The current research investigates the conductivity σ AC of glassy As4Ge24Te72 thin films through a combined experimental and theoretical approach. It sheds light on the complex relationship between σ AC , frequency, temperature, and film thickness. Analysis reveals that σ AC follows ω r law and exhibits hopping behavior. To effectively model and predict σ AC , an adaptive neuro-fuzzy inference system (ANFIS) is utilized. The ANFIS model successfully simulates the experimental data, showing high accuracy. Additionally, the prediction of experimentally measured values of σ AC is processed as a testing step and provides acceptable results. That enables ANFIS to enlarge the scale and complete the missing parts in the trained datasets by predictions for unmeasured σ AC values. The ANFIS network was built using MATLAB-R2017a. It consists of two inputs (frequency and temperature) and one output (AC conductivity). The precision is confirmed by calculating the modeling errors across the training process. The mean absolute errors MAE through the modeling process are not exceeding 10−5. This average error value demonstrates the model’s effective ability. The alignment between experimental and ANFIS modeling results open up exciting possibilities for future research. By predicting the properties of the understudy material using ANFIS model, we can pave the way for more compact and efficient electrical devices.

A fluorescence based dual sensor for Zn2+ and PO43- and the application of soft computing methods to predict machine learning outcomes.

A phenolphthalein-based Schiff base, 3,3-bis-{4-hydroxy-3-[(pyridine-2-ylmethylimino)-methyl]-phenyl}-3H-isobenzofuran-1-one (PAP), has been synthesized and used for selective fluorescence 'turn on' and 'turn off' sensing of Zn2+ and PO43- respectively. The limit of detection using the 3σ method for Zn2+ is found to be 19.3 nM and that for PO43- is 8.3 μM. The sensing mechanism of PAP for Zn2+ ions has been explained by 1H NMR, 13C NMR, TRPL, ESI-MS, FT-IR, and DFT based calculations. Taking advantage of this fluorescence 'on-off' behavior of PAP in the sequential presence of Zn2+ and PO43- a two input fuzzy logic (FL) operation has been constructed. The chemosensor PAP can thus act as a metal ion and anion responsive molecular switch, and its corresponding emission intensity is used to mimic numerous FL functions. To replace various expensive, time-consuming experimental procedures, we implemented machine learning soft computing tools, such as fuzzy-logic, artificial neural networks (ANNs), and adaptive neuro-fuzzy inference systems (ANFIS), to correlate as well as predict the fluorescence intensity in the presence of any equivalent ratio of Zn2+ and PO43-. The statistical performance measures (MSE and RMSE, for example) show that the projected values of the cation and anion sensing data by the ANFIS network are the best and closer to the experimental values.

Multi-characteristic optimization and modeling analysis of electrocoagulation treatment of abattoir wastewater using iron electrode pairs

Multi-characteristic optimization and modeling analysis of electrocoagulation (EC) treatment of abattoir wastewater (AWW) using iron‑iron electrodes are reported. Response Surface Methodology (RSM) and Artificial Intelligent (AI) modeling tools viz. Artificial Neural Network (ANN) and Adaptive Neuro-Fuzzy Inference System (ANFIS) were used for the modeling while the numerical method of RSM and RSM-based Genetic Algorithm (RSM-GA) were used to optimize the response variable. The independent variables were pH, current intensity, electrolysis time, settling time, and temperature while the dependent variable was percentage turbidity removal. Based on high determination coefficient (R2) and low standard deviation (SD) values, the quadratic model was selected from the ANOVA of the RSM model. For the ANN model, the number of neurons were varied between 2 and 10 but based on low MSE and high R2 values, 10 was selected as the optimum number of neurons. The ANFIS network used the triangular input membership function (trimf) with 100 epochs. The three models depicted linear adequacy with respective R2 values of 0.9978, 0.9995, and 0.8285 for RSM, ANFIS, and ANN respectively. However, the performance error indices indicated that the prediction accuracy of the three models followed the order — ANFIS > ANN > RSM. Further, the validated optimization results produced 94.74 % and 97.42 % turbidity removal for the RSM and RSM-GA optimization methods respectively. This proved that RSM, ANN, and ANFIS can reliably be used to model the EC treatment of AWW using FeFe electrodes; while the response variable can successfully be optimized by RSM and RSM-GA methods.

Modeling of dielectric behavior of polymers nanocomposites using adaptive neuro-fuzzy inference system (ANFIS)

The research presents a theoretical study on modeling of dielectric properties of polymer nanocomposites based on adaptive neuro-fuzzy inference system (ANFIS). The research contributes a better utilizing of ANFIS model in the prediction of dielectric behavior of polymers nanocomposites. In this respect, three different samples are trained (Mg1−xCuxO/PMMA, PPy-DBSA-Y2O3 and PVC/PEMA with [Zn(CF3SO3)2]). Inputs are obtained from earlier experimental studies. ANFIS Takagi–Sugeno type is trained. The model is applied based on weighted average as a defuzzification method. The optimal network structures, which produce the most acceptable results, are implemented in MATLAB. Six ANFIS networks are trained to simulate and predict dielectric permittivity and dielectric loss in terms of nanocomposite weight % (0–0.2%, 0–8% and 10–30% for each sample, respectively) and frequency (10−2–103 kHz). ANFIS simulation results are very close to their targets. Predictions of dielectric properties at nanocomposite weights % that are measured experimentally as a testing step and predictions of other values that are not implicated in the experimental data extent are achieved. Also, predictions of individual points are processed using ANFIS rule viewer. It is found that ANFIS predictions provide excellent results. Three-dimensional illustrations that represent the mapping from frequency and nanocomposite weight% to the dielectric permittivity and dielectric loss are obtained using MATLAB surface viewer. Histogram error plot is obtained to indicate the degree of noisy. Mean error, mean squared error, root-mean-squared error and standard division error are calculated. Their values improve the efficiency of the modeling process. A key goal of this paper is to develop a mechanism to predict the dielectric properties of polymers nanocomposites. In accordance with the modeling results, the ANFIS technique achieves the purpose. It can also form a great link between practical and theoretical domains.

ANFIS Based Thermal Estimation of Ultradeep Submicron Digital Circuit Design

In this paper, the use of the Adaptive Neuro Fuzzy Inference System (ANFIS) to model the CMOS inverter is discussed as a tool for developing and simulating CMOS logic circuits at the ultradeep submicron technology node of 22nm. The ANFIS structures are built and trained using MATLAB software. The ANFIS network was trained using data obtained from the analytical model (at 298.15K and 398.15K). For training, two methodologies are used: a hybrid learning method based on back-propagation and least-squares estimation, and back-propagation. The effect of the ANFIS model's structure on the accuracy and performance of the CMOS inverter has also been investigated. Further, simulation through HSPICE using (Predictive Technology Model) PTM nominal parameters has been done to compare with ANFIS (trained using an analytical model) results. The comparison of ANFIS and HSPICE suggests the ANFIS modelling procedure's practicality and correctness. The findings demonstrate that the ANFIS simulation is significantly faster and more comparable than the HSPICE simulation and that it can be easily integrated into software tools for designing and simulating complicated CMOS logic circuits.

Modeling and optimization of nano-rod plasmonic sensor by adaptive neuro fuzzy inference system (ANFIS)

Predicting the behavior of photonic and Plasmon nanostructures has always been an attractive topic for photonic sciences researchers. One of the methods for estimating the behavior of devices in engineering sciences is adaptive neuro fuzzy inference system. Modeling and designing of photonic and plasmonic nanostructures and their optimization strongly relies on the timing of unstructured electromagnetic response simulations and optimal design requires a high number of simulations and if the optimal answer is reached, other simulation results will be wasted. Deep learning method which has been created by ANFIS networks provides a powerful and efficient device for creating accurate relationship between plasmonic geometric parameters and resonance spectrums. Millions of different nanostructures can be obtained without the need for any costly simulations and it costs only one investment to obtain training data. In this study, we use of deep learning based on ANFIS for predicting optic sensors’ answer by using plasmonic nano rods. In addition to prediction, this method can approximate a special answer based on photonic structure. In this study, ANFIS networks, at first, model a 5*5 array of plasmonic nanostructures based on obtained experimental data from optic absorption. In the following, based on obtained model, it designs nanoparticles geometry so that we achieve a relatively narrow absorption spectrum with high sensitivity in relation to change air refractive index. This method can be applied for other similar types of nano-photonic systems which can help to destroy simulation procedure and acceleration photonic sensor design trend.

Processing watermelon waste using Saccharomyces cerevisiae yeast and the fermentation method for bioethanol production

AbstractBioethanol is a source of energy for the future when petroleum is going to finish. The aim of this study is to assess bioethanol production from watermelon waste using Saccharomyces cerevisiae (S. cerevisiae) yeast and fermentation method. The test was replicated three times in 35 hr at three different fermenter agitator speeds and three levels of yeast. The results showed that about 35.5 g of bioethanol was obtained from each kilogram of watermelon. Investigating different variables shows that the fermenter agitator speed of 120 rpm and the yeast amount of 5 g could lead to the best yield in the process of fermenting watermelon waste for the purpose of producing bioethanol. The results of the evaluating the artificial neural network (ANN) model and adaptive neuro‐fuzzy inference system (ANFIS) in predicting bioethanol production from watermelon waste with the highest coefficient of determination (R2) were 0.9895 and 0.9993, respectively. These results indicate that ANNs and ANFIS are effective in predicting bioethanol production from watermelon waste.Practical ApplicationsAdvanced, green, clean, and sustainable processing technologies to reduce watermelon waste. Fermentation of agricultural waste in the shortest time (30 hr). Investigating important factors in waste fermentation process. The performance of ANN and ANFIS network in predicting bioethanol production.

Saturation magnetization parameters by adaptive neuro-fuzzy technique

The method of ANFIS (adaptive neuro fuzzy inference system) was applied to the data resulting from these measurements. The ANFIS process for variable selection was implemented in order to detect the predominant variables affecting the saturation magnetization prediction. This process includes several ways to discover a subset of the total set of recorded parameters, showing good predictive capability. The correlation of the involved parameters with the growth process was examined by employing the central composite design method through designating set up experiments that will determine the interaction of the variables. The vibrating sample magnetometer (VSM) was used to confirm the statistical analysis. The ANFIS network was used to perform a variable search. Then, it was used to determine how 4 parameters (pH, Temperature (∘C), Time (min), Alkali rate of addition (mL.s−1)) influence saturation magnetization prediction. The results indicated that pH is the most influential to sugarcane growth prediction, and the best predictor of accuracy. The response obtained by VSM suggests that the saturation magnetization of nanomagnetite particles can be controlled by restricting the effective parameters.

Observer design for a nano-positioning system using neural, fuzzy and ANFIS networks

This paper focuses on the observer design for a 2D nano-positioner. In order to position the stage with a desired accuracy, it is required to adjust the stage displacements with a closed-loop control system. Since displacement and velocity of the main stage are not measured directly in the designed nano-positioning system, some observers should be designed to estimate these state variables using data provided by measurable variables. To this end, three different observers were designed based on neural, fuzzy and adaptive neuro fuzzy inference system (ANFIS) networks. With the purpose of obtaining data for training the observer model, a reference model is required. For this reason, the mechanism was modelled in COMSOL. The results show that among the designed observers, ANFIS can estimate the system states with higher accuracy compared to the two other observers. Currently, the position of the stage in the commercial XY nano-positioners is measured using two relatively high-cost capacity sensors that need driver circuits, while based on the observation scheme proposed in this paper, the complexity and cost of the nano-positioning systems can be reduced by using strain gauge type sensors mounted on piezo-actuators.

Potential of adaptive neuro-fuzzy methodology for investigation of heat transfer enhancement of a minichannel heat sink

In this paper, the potential of adaptive neuro fuzzy inference system (ANFIS) was appraised for investigation of thermal performances of a minichannel heat sink for cooling of electronics using nanofluid coolant instead of pure water. The process, which simulates the thermal performances with ANFIS network, was constructed. The developed ANFIS network was with three neurons in the input layer, and one neuron (thermal performances) in the output layer. Since there are three thermal performances investigated, three ANFIS networks are created. The inputs were flow rate, as well as the Reynolds number. The third input represents the nanofluid concentration. The Al2O3–H2O nanofluid including the volume fraction was used as a coolant. Obtained experimental results showed the higher improvement of the thermal performances using nanofluid instead of pure distilled water. ANFIS results show that an improvement in predictive accuracy and capability of generalization can be achieved by the ANFIS approach for heat sink base temperature prediction.

Time‐Dependent Reliability Analysis of RC Deep Beams considering Linear/Nonlinear Creep and Shrinkage Using ANFIS Network and MCS

This paper comprehensively studies the effects of linear and nonlinear creep and shrinkage on shear strength and deformation of reinforced concrete (RC) deep beams through a time‐dependent reliability analysis framework. The three‐dimensional finite element‐based program of ABAQUS was used to model RC deep beams. A viscoelastic approach was adopted to model linear creep and shrinkage by a user‐defined material model developed and implemented in user subroutines UMAT and UEXPAN. The software was initially examined using two experimental results for short/long‐term behavior of shallow and deep concrete beams. In the nonlinear range, creep was taken into account by the affinity hypothesis theorem to consider the effect of high‐level sustained loading. Due to the complicated and time‐consuming nature of the finite element method (FEM), adaptive neuro‐fuzzy inference system (ANFIS) and Monte Carlo simulation (MCS) replaced these complex analyses. Finally, based on the numerical results obtained from the analyses of case study samples, it was concluded that, in a serviceability limit state, for RC deep beams, the probability of failure was reduced to less than one‐fifth that for the shallow beams. On the contrary, in a strength limit state, a safety factor of about 1.7∼1.8 could be considered for the effect of sustained high‐level loading on the shear strength of RC deep beams.

150. DETECTION OF EPILEPSY USING ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM

This study presents the use of Adaptive Neuro-Fuzzy Inference System (ANFIS) for classification of the EEG signals. The data consists of two types of EEG signals, i.e. epileptic patients during epilepsy and healthy patients when their eyes are open. The proposed algorithm has several steps. First, in order to remove the artefacts (filter the signals) we use band-pass Finite Impulse Response (FIR) filtering with the Hamming window. Feature extraction is made in the second step, using Discrete Wavelet Transform (DWT) and statistical analysis. In this way we reduce the dimensionality of the input data, lately used as input parameters in the ANFIS network. ANFIS model learns how to classify the EEG signal, through the standard hybrid learning algorithm. We use special form of ANFIS model, which depending on the number of inputs, splits the model into appropriate number of substructures (sub-ANFIS models). ANFIS model was evaluated in terms of training performance and classification accuracies. From the simulation results it was concluded that the proposed algorithm has good potentials in classifying the EEG signals.

The effect of ultrasound pre‐treatment on quality, drying, and thermodynamic attributes of almond kernel under convective dryer using ANNs and ANFIS network

AbstractIn this study, different drying conditions were investigated on quality and thermodynamic properties of almond kernel. Experiments were performed using a convection dryer with ultrasound pretreatment in 40, 50, 60, and 70 °C air temperature, 1 m/s air velocity, and duration of ultrasonic pre‐treatment of 0 min (for control sample), 10, 20, and 40 min. The drying kinetic of the almond kernel was estimated by 15 mathematical models. Furthermore, Artificial Neural Networks (ANNs) and Adaptive Neuro‐Fuzzy Inference Systems (ANFIS) were applied to fit the experimental data on the thin layer drying. The lowest and highest values of the effective moisture diffusivity (Deff) was 1.81 × 10−9 and 9.70 × 10−9 m2/s, respectively. Activation energy (Ea) of the samples was obtained between 26.35 and 36.44 kJ/mol. The highest and lowest values of specific energy consumption (SEC) were calculated 561.72 and 169.88 kW hr/kg, respectively. Maximum (13.14%) and the minimum (7.1%) values of shrinkage were achieved at air temperatures of 70 and 40 °C, respectively. The color changing of dried samples was obtained between 9.14 and 17.96. Furthermore, results revealed that the ANFIS model had the high ability to predict the moisture ratio (R2 = 0.9998 and MSE = 0.0003) during drying. As a result, ANFIS model has the highest ability to evaluate all output as compared with other models and ANNs method.Practical applicationsAlgorithms are modern methods that have been successfully applied to solve the various problems and modeling in engineering and science. Drying is one of the oldest procedures to preserve the food quality. Reduction of moisture content to a certain value can be caused to decay and minimize the microbiological activity and deteriorating chemical reactions in agricultural products, respectively. Determination of almond drying process under convective with ultrasound pre‐treatment dryer in terms of desirable thermal properties (effective moisture diffusivity and energy consumption) provides the high‐quality products. Furthermore, this research can be able to provide a technical basis for almond drying and the related equipment designing.

Power production prediction of wind turbines using a fusion of MLP and ANFIS networks

Access to accurate power production prediction of a wind turbine in future hours enables operators to detect possible underperformance and anomalies in advance. This may enable more proactive and strategic operations optimisation. This study examines common Supervisory Control And Data Acquisition (SCADA) data over a period of 20 months for 21 pitch regulated 2.3 MW turbines. In this study, an algorithm is proposed to impute values of data that are missing, out-of-range, or outliers. It is shown that an appropriate combination of a decision tree and mean value for imputation can improve the data analysis and prediction performance by the creation of a smoother dataset. In addition, principal component analysis is employed to extract parameters with power production influence based on all available signals in the SCADA data. Then, a new data fusion technique is applied, combining dynamic multilayer perceptron (MLP) and adaptive neuro-fuzzy inference system (ANFIS) networks to predict future performance of wind turbines. This prediction is made on a scale of one-hour intervals. This novel combination of feature extraction, imputation, and MLP/ANFIS fusion performs well with favourably low prediction error levels. Thus, such an approach may be a valuable tool for turbine power production prediction.

Semi-active seismic control of buildings using MR damper and adaptive neural-fuzzy intelligent controller optimized with genetic algorithm

This research presents designing a control system to reduce seismic responses of structures. Semi-active control of a magnetorheological (MR) damper is used to improve seismic behavior of a 3-story building implementing neural-fuzzy controller made of adaptive neuro-fuzzy inference system (ANFIS) to determine damper input voltage. Both premise and consequent parameters of fuzzy membership and output functions of ANFIS have the ability for training and improvement but most researchers have focused on just consequent parameters. In order to optimize the controller performance, an approach is proposed in this paper where both premise and consequent parameters of fuzzy functions in an ANFIS network are adjusted simultaneously by genetic algorithm (GA). In order to assess the effectiveness of the designed control system, its function is numerically studied on a benchmark 3-story building and is compared to those of a neural network predictive control (NNPC) algorithm, linear quadratic Gaussian (LQG) and clipped optimal control (COC) systems in terms of seismic performance. The results showed desirable performance of the (ANFIS +GA + membership functions + result function) ANFIS–GA–MFR controller in considerably reducing the structure responses under different earthquakes. The proposed controller showed 30 and 39% reductions in peak story drift (J1) and normed story drift (J4) respectively compared to the NNPC controller, 32 and 44% reductions in J1 and J4 respectively compared to the LQG controller, and 27 and 38% reductions in J1 and J4 respectively compared to the COC controller. The proposed controller effectively reduced acceleration and base shear level compared to the uncontrolled state and had a performance relatively similar to those of three other controllers – for instance, it reduced the maximum level acceleration (J2) 10% higher than COC. Also, the results showed that the ANFIS–GA–MFR controller has more efficiency than the basic ANFIS controller, on average about 20%.

ANFIS multi criteria decision making for overseas construction projects: a methodology

A critical part when a company targeting a foreign market is how to make a better decision in connection with potential project selection. Since different attributes of information are often incomplete, imprecise and ill-defined in overseas projects selection, the process of decision making by relying on the experiences and intuition is a risky attitude. This paper aims to demonstrate a decision support method in deciding overseas construction projects (OCPs). An Adaptive Neuro-Fuzzy Inference System (ANFIS), the amalgamation of Neural Network and Fuzzy Theory, was used as decision support tool to decide to go or not go on OCPs. Root mean square error (RMSE) and coefficient of correlation (R) were employed to identify the ANFIS system indicating an optimum and efficient result. The optimum result was obtained from ANFIS network with two input membership functions, Gaussian membership function (gaussmf) and hybrid optimization method. The result shows that ANFIS may help the decision-making process for go/not go decision in OCPs.

Volume fraction determination of the annular three-phase flow of gas-oil-water using adaptive neuro-fuzzy inference system

The use of adaptive neuro-fuzzy inference system (ANFIS) has been reported for predicting the volume fractions in a gas–oil–water multiphase system. In fact, the volume fractions in the annular three-phase flow are measured based on a dual energy metering system consisting of $$^{152}\hbox {Eu}$$ and $$^{137}\hbox {Cs}$$ and one NaI detector using ANFIS. Since the summation of volume fractions is constant, therefore ANFIS must predict only two volume fractions. In this study, three ANFIS networks are applied. The first is utilized to predict the gas and water volume fractions. The next one is applied to predict the gas and oil, and the last one is used to predict the water and oil volume fractions. In the next step, ANFIS networks must be trained based on numerically obtained data from MCNP-X code. Then, the average testing errors of these three networks are computed and compared. The network with the least error has been selected as the best predictor model.

Prediction and optimization of hydrogen yield and energy conversion efficiency in a non-catalytic filtration combustion reactor for jet A and butanol fuels

Hydrogen production is one of main subjects in fuel cells. The traditional method of synthesis gas production is based on fuel reforming using catalysts. The main problem of these methods is sensitivity and fast degradation of catalysts especially when fuels with high sulfur content are used. A new technique for hydrogen production is fuel-reforming using non-catalytic filtration combustion in porous media reactors. Various experimental works have been carried out to increase hydrogen production under different operating conditions such as inlet fuel velocity and equivalence ratio. First, we investigated the ability of adaptive neuro fuzzy inference system (ANFIS) for predicting the filtration combustion characteristics. Four distinct ANFIS models were developed for estimating the hydrogen yield and energy conversion efficiency for fuels of jet A and butanol. Eight different membership functions of dsigmf, gauss2mf, gaussmf, gbellmf, pimf, psigmf, trapmf and trimf were tested for training of the ANFIS networks. The results showed that the RMSE of the best developed ANFIS models for estimating of the hydrogen yield of jet fuel, hydrogen yield of butanol, conversion efficiency of jet fuel and conversion efficiency of butanol were 1.399, 1.213, 0.508 and 2.191, respectively. Moreover the R2 values of 0.998, 0.998, 0.999 and 0.999 were obtained for predicting the above mentioned variables, respectively. In the second step, a novel algorithm based on imperialist competitive algorithm (ICA) was used for optimization of hydrogen yield and energy efficiency. The maximum value of hydrogen yield and energy efficiency was 50.46% and 67.88% for jet A and 47.27% and 96.93% for butanol, respectively. The results showed that the imperialist competitive algorithm is an efficient and powerful algorithm to optimize combustion processes.

Modeling of shear wave velocity in limestone by soft computing methods

The main purpose of current study is development of an intelligent model for estimation of shear wave velocity in limestone. Shear wave velocity is one of the most important rock dynamic parameters. Because rocks have complicated structure, direct determination of this parameter takes time, spends expenditure and requires accuracy. On the other hand, there are no precise equations for indirect determination of it; most of them are empirical. By using data sets of several dams of Iran and neuro-genetic, adaptive neuro-fuzzy inference system (ANFIS), and gene expression programming (GEP) methods, models are rendered for prediction of shear wave velocity in limestone. Totally, 516 sets of data has been used for modeling. From these data sets, 413 ones have been utilized for building the intelligent model, and 103 have been used for their performance evaluation. Compressional wave velocity (Vp), density (γ) and porosity (n), were considered as input parameters. Respectively, the amount of R for neuro-genetic and ANFIS networks was 0.959 and 0.963. In addition, by using GEP, three equations are obtained; the best of them has 0.958R. ANFIS shows the best prediction results, whereas GEP indicates proper equations. Because these equations have accuracy, they could be used for prediction of shear wave velocity for limestone in the future.

Artificial Neural Network and Adaptive Neuro-Fuzzy Interface System Modeling of Supercritical CO2 Extraction of Glycyrrhizic Acid from Glycyrrhiza glabra L

Abstract In this study, the extraction of Glycyrrhizic acid (GA) from Glycyrrhiza glabra (licorice) root was investigated by Soxhlet extraction and modified supercritical CO2 with water as co-solvents and 30 min of static extraction time. The high performance liquid chromatography (HPLC) was used to identify and quantitatively determine the amount of extracted GA recovery of supercritical CO2 extraction of GA. The extraction recovery was modeled by adaptive neuro-fuzzy inference system (ANFIS) and artificial neural network (ANN). Different ANFIS networks (by changing the type of membership functions) were compared with evaluation of networks accuracy in GA recovery prediction and subsequently the suitable network was determined. A three-layer artificial neural network was also developed for modeling of GA extraction from licorice plant root. In this regard, different networks (by changing the number of neurons in the hidden layer and algorithm of network training) were compared with evaluation of networks accuracy in extraction recovery prediction. One-step secant back propagation algorithm with six neurons in hidden layer was found to be the most suitable network and the coefficient of determination (R2) was 98.5 %. Gaussian combination membership function (gauss2mf) using 2 membership function to each input was obtained to be optimum ANFIS architecture with mean square error (MSE) of 0.05,0.17 and 0.07 for training, testing and checking data, respectively.