Artificial Neural Network Modeling Technique Research Articles

RSM and ANN Modeling Techniques to Forecast How Various Parameters will Affect the Improvement of Electronics Cooling using Radial Heat Sink.

This research provides the mathematical modeling for temperature difference for natural convection using Response Surface Methodology (RSM) and Artificial Neural Network (ANN) based modelling. Length of fin (L), height of fin (H), number of fins (N) and the heat input (Q) for the Radial heat sink are the parameters selected under natural convection heat transfer. Looking at the pattern of the data, feed forward back propagation type neural network is chosen. The RSM mathematical model of temperature difference is used to compare the performance of the created ANN models. ANN Simulations proved to be successful in terms of agreement with actual values of experimentation. ANN simulations perform accurate to validate the experimental results and the results obtained from the RSM for the output under natural convection. The optimum values for the dimensional parameters namely length of fin, height of the fin, number of fins are obtained by RSM method. Also the optimum operating parameter that is heat input for minimum temperature difference is obtained.

Artificial Neural Network Modeling Techniques for Drying Kinetics of Citrus medica Fruit during the Freeze-Drying Process

The main objective of this study is to analyze the drying kinetics of Citrus medica by using the freeze-drying method at various thicknesses (3, 5, and 7 mm) and cabin pressures (0.008, 0.010, and 0.012 mbar). Additionally, the study aims to evaluate the efficacy of an artificial neural network (ANN) in estimating crucial parameters like dimensionless mass loss ratio (MR), moisture content, and drying rate. Feedforward multilayer perceptron (MLP) neural network architecture was employed to model the freeze-drying process of Citrus medica. The ANN architecture was trained using a dataset covering various drying conditions and product characteristics. The training process, including hyperparameter optimization, is detailed and the performance of the ANN is evaluated using robust metrics such as RMSE and R². As a result of comparing the experimental MR with the predicted MR of the ANN modeling created by considering various product thicknesses and cabin pressures, the R² was found to be 0.998 and the RMSE was 0.010574. Additionally, color change, water activity, and effective moisture diffusivity were examined in this study. As a result of the experiments, the color change in freeze-dried Citrus medica fruits was between 6.9 ± 0.2 and 21.0 ± 0.6, water activity was between 0.4086 ± 0.0104 and 0.5925 ± 0.0064, effective moisture diffusivity was between 4.19 × and 21.4 × , respectively. In freeze-drying experiments conducted at various cabin pressures, it was observed that increasing the slice thickness of Citrus medica fruit resulted in longer drying times, higher water activity, greater color changes, and increased effective moisture diffusivity. By applying the experimental data to mathematical models and an ANN, the optimal process conditions were determined. The results of this study indicate that ANNs can potentially be applied to characterize the freeze-drying process of Citrus medica.

Optimal pretreatment of plantain peel waste valorization for biogas production: Insights into neural network modeling and kinetic analysis

This work proposed a model for the substrate treatment stage of biogas production process in an anaerobic digestion system. Adaptive neuro-fuzzy inference system (ANFIS), response surface method (RSM), and artificial neural network (ANN) were comparatively used in the simulation and modeling of the treatment process for improved biogas yield. Waste plantain peels were pretreated and used as substrate. FTIR and SEM results revealed that the pretreatment improved the substrate's desirable qualities. The amount of biogas yield was controlled by time, NaOH concentration, and temperature of the substrate pretreatment. Optimum pretreatment conditions obtained were a temperature of 102.7 °C, time of 31.7 min and NaOH concentration of 0.125 N. RSM, ANN, and ANFIS modeling techniques were proficient in simulating the biogas production, as evidenced by high R2values of 0.9281, 0.9850, and 0.9852, respectively. Furthermore, the values of the calculated error terms such as RMSE (RSM = 0.04799, ANN = 0.00969, and ANFIS = 0.00587) and HYBRID (RSM = 18.556, ANN = 0.803, and ANFIS = 0.0447) were low, indicating a satisfactory correlation between experimental and predicted values. Scrubbing of the biogas with caustic soda and activated charcoal increased the methane content to 94 %. The kinetics of the cumulative biogas yield were best fit with the Logistics and Modified Logistics models. The low C/N ratio in addition to the presence of potassium, nitrogen, and phosphorus suggested that the spent plantain peel slurry can be utilized as an agricultural fertilizer in crop production. The observations of this study therefore recommends the pre-treatment of biodigestion substrates as a key means to enhance beneficiation of methane production.

Uncertainty in Artificial Neural Network Models: Monte-Carlo Simulations Beyond the GUM Boundaries

ABSTRACT In this study, the accuracy of the artificial neural network (ANN) was assessed considering the uncertainties associated with the randomness of the data and the lack of learning. The Monte-Carlo algorithm was applied to simulate the randomness of the input variables and evaluate the output distribution. It has been shown that under certain conditions, the GUM framework for uncertainty evaluation may completely fail. The ANN modeling technique can be used as an alternative method for estimating the expectation value and evaluating the associated uncertainty. Furthermore, unlike the GUM and Monte-Carlo frameworks, the ANN models do not require mathematical expressions between the input and output variables. On the other hand, owing to the uncertainty associated with the lack of learning, the ANN model may produce unrealistic results, even if a global minimum is approached. This behavior is explained by Bayesian theory which assumes that the output values generated by various runs are normally distributed at each target. This may lead to an unrealistic output when the overall distribution of the target values has a different distribution than that presumed by Bayesian theory. To minimize this drawback, the ANN model output should be calculated from sufficiently large repeated runs with new starting values of the weights and biases.

Synthesis of drug carrier smart ferrogels and application of artificial neural network in modeling their doxorubicin release behavior under alternating magnetic fields

AbstractThe development of magnetic field‐sensitive smart drug delivery systems with superior properties has become an area of increasing academic and industrial importance. In this study, smart poly(NIPAAm‐co‐VSA)‐rGO/Fe3O4 ferrogels with varying concentrations of reduced graphene oxide (rGO) were synthesized. Fe3O4 nanoparticles loaded ferrogels were obtained by the in situ reduction of Fe ions. The morphologic, structural, and magnetic properties of ferrogel systems were characterized. Doxorubicin (DOX) was loaded to the synthesized ferrogels by solution impregnation method and their heating and drug release behavior over time under alternating magnetic field AMFs of 1.37, 1.64, and 1.91 millitesla (mT) were investigated. The drug loading and releasing characteristics of the ferrogel series were calculated. When the experimental results were examined, it was determined that the amount of rGO in the structure of the developed ferrogel systems had a very high effect on the magnetic, heating, and drug loading‐release characteristics of the ferrogels. To modeling their multivariable DOX release behavior, artificial neural network modeling technique was used. Calculated model performance parameters have shown that this developed artificial intelligence technique has great success in modeling complex and nonlinear DOX release behaviors.

Optimization and prediction of incremental sheet forming parameters of Titanium grade 5 sheet using a response surface methodology and artificial neural network

Single point incremental forming (SPIF) is an advanced, flexible, and cost-effective approach for producing complicated sheet metal objects quickly. Because of its low equipment cost, the process is best suited toward low or medium quantity batch production as the conventional stamping method remains cost-effective only for mass manufacture. During SPIF formation of titanium grade 5 materials, a response surface methodology and artificial neural network (ANN) model was created to optimize and estimate wall angle (Ømax) and average surface roughness ( Ra). The ANN model is developed using feed forward back propagation networks. Various combinations of transfer functions and number of neurons were used to create the ANNs (3- n-1, 3- n-2). The confirmation runs have been used to ensure that the ANN anticipated and practical findings have been in agreement. The generated ANN model (3- n-1) was capable of accurately forecasting the results of the experiment, with an overall R-value and Mean Square Error (MSE) of 0.99987 and 0.010905 for Ra, and 0.99999 and 0.00700 for wall angle. The best 3- n-2 models has an average R-value of 0.99992 and MSE of 0.05532, respectively. As a consequence of rapid ANN modeling technique, it became discovered that technical effort inside the SPIF process could be decreased.

Towards Predictive Maintenance of a Heavy-Duty Gas Turbine A New Hybrid Intelligent Methodology for Performance Simulation

As the increasing demands of global clean power for the main purpose of lowering environmental pollution, heavy-duty gas turbines are playing an increasingly important role in energy fields because of their low emission, high thermal efficiency and flexible start-up capacity. Accurate modeling and simulation of the turbine performance is extremely needed to precisely design a large-watt industrial heavy-duty gas turbine, which is still very challenging due to the high nonlinearity, system complexity, varying conditions, and strong coupling interaction of high-dimension parameters under the harsh operation environment of the turbine. In order to improve the accuracy and efficiency of the performance simulation model for a heavy-duty gas turbine at various operating conditions, this paper presents a new hybrid intelligent methodology to adeptly integrate system thermodynamic balance mechanism with multivariate data by seamlessly combing advanced signal processing, machine learning and artificial neural network modeling techniques. The thermodynamic model of a complicated single-shaft gas turbine is first created based on the balances of both flow and power in various subsystems including inlet, compressor, turbine, combustor and exhaust. The characteristic curves of compressor and turbine are utilized to accurately represent the physical mechanism and effectively simulate the high nonlinear behaviors of subsystems. Wavelet signal processing and machine learning based feature extraction are employed to preprocess the multivariate raw data of the turbine. Multilayer artificial neural network models are explored to efficiently simulate the start-up transient process of the turbine, thus improve the simulation efficiency and accuracy of the complicated system. Multivariate data collected from a real-world industrial heavy-duty gas turbine is employed to illustrate the effectiveness and feasibility of the proposed methodology.

Optimization of Wood Particleboard Drilling Operating Parameters by Means of the Artificial Neural Network Modeling Technique and Response Surface Methodology

Drilling is one of the oldest and most important methods of processing wood and wood-based materials. Knowing the optimum value of factors that affect the drilling process could lead both to high-quality furniture and low-energy consumption during the manufacturing process. In this work, the artificial neural network modeling technique and response surface methodology were employed to reveal the optimum value of selected factors, namely, drill tip angle, tooth bite, and drill type of the delamination factor at the inlet and outlet, thrust force, and drilling torque. The data set that was used in this work to develop and validate the ANN models was collected from the literature. The results showed that the developed ANN models could reasonably predict the analyzed responses. By using these models and the response surface methodology, the optimum values of analyzed factors were revealed. Moreover, the influences of selected factors on the drilling process of wood particleboards were analyzed.

Application of statistical and soft computational techniques in machining of Nickel based supper-alloy using cryogenically treated tools for estimation of surface roughness

ABSTRACT In this paper, surface roughness prediction models are developed for turning of Inconel 718 using untreated and cryogenically treated inserts by using Dimensional Analysis, Response Surface Methodology (RSM) and Artificial Neural Network (ANN). Performance of untreated and treated tools is analysed using SEM, Energy-dispersive X-ray analysis, Vicker hardness test and electrical conductivity. For the established surface roughness models by dimensional analysis, RSM and ANN, the mean absolute errors for confirmation tests are 5.32%, 8.28% and 4.15% for untreated inserts and 4.95%, 6.01% and 4.20% for treated inserts, respectively. The effect of cutting parameters on surface roughness is analysed using the main effect plot and 3D surface plots. Based on correlation coefficient (R2 ) values, ANN modelling technique (R2 = 99.68%) is more accurate for predicting surface roughness. Thus, it can be an effective tool for analysing machining responses. The study also noted that while cutting at v= 60 m/min, f= 0.1 mm/rev and d= 0.5 mm, surface roughness and flank wear values are 0.5 µm and 0.45 µm and 0.777 mm and 0.627 mm for untreated and treated inserts, respectively. The use of treated tools resulted in 10% and 19% improvement in surface quality and tool life than the untreated tools.

A Model for Stock Market Value Forecasting using Ensemble Artificial Neural Network

Artificial Neural Network (ANN) is a model used in capturing linear and non-linear relationship of input and output data. Its usage has been predominant in the prediction and forecasting market time series. However, there has been low bias and high variance issues associated with ANN models such as the simple multi-layer perceptron model. This usually happens when training large dataset. The objective of this work was to develop an efficient forecasting model using Ensemble ANN to unravel the market mysteries for accurate decision on investment. This paper employed the Ensemble ANN modeling technique to tackle the high variations in stock market training dataset faced when using a simple multi-layer perceptron model by using the theory of ensemble averaging. The Ensemble ANN model was developed and implemented using NeurophStudio and Java programming language, then trained and tested using daily data of stock market prices from various banks, for a period of 497 days. The methodology adopted to achieve this task is the agile methodology. The output of the proposed predictive model was compared with four traditional neural network multilayer perceptron algorithms, and outperformed the traditional neural network multilayer perceptron algorithms. The proposed model gave an average to best predictive error for any day when compared with the other four traditional models.

A Machine Learning-Based QSAR Model for Quinolines Tested As Corrosion Inhibitors of Mild Steel in HCl Solution

The investigation of organic compounds as potential inhibitors of metal corrosion using several experimental and/or computational techniques has been well established in literature. Recent advances in the development of corrosion inhibitors has led to the exploration of quantitative structure activity relationship (QSAR) as a worthy alternative to the rigorous, sometimes expensive and time-consuming experimental/computational procedures. QSAR is a robust computational method that generates mathematical models that maps the structural features of organic compounds with their inhibition performances. The obtained mathematical models can then be further used to predict the chemical activity of new and untested organic molecules. Although linear and non-linear methods have been reported in developing QSAR models, only few works have explored the use of state-of-the-art artificial neural network (ANN). This study reports the development of QSAR models for sixty quinolines investigated as corrosion inhibitors for mild steel corrosion in 1 M HCl using both linear and non-linear modelling techniques. Density functional theory (DFT) and Dragon 7 software were used to generate molecular descriptors for model development and were reduced to a significant few using feature selection tools. The significant molecular descriptors were employed in model building using conventional multiple linear regression (MLR) and ANN modelling techniques. The developed models were evaluated using several statistical performance metrics and the ANN-based model was found to produce the best correlation between the predicted and experimental inhibition efficiencies. Furthermore, nine new quinoline molecules were theoretically designed and their inhibition efficiencies were predicted using the best developed ANN model.

Identifying most influencing input parameters for predicting chloride concentration in groundwater using an ANN approach

Assessment of groundwater quality at a specific location is an important step to provide adequate information about water management and sustainable development. Several variables affect groundwater salinity, expressed by chloride concentration, prediction; therefore, identification of the most significant parameters for accurate prediction is an important research area. In the present study, artificial neural network (ANN) models with various combinations of input parameters were developed to determine the most significant parameters that influence chloride concentration prediction. To achieve this, the variables affecting chloride concentration (recharge rate (RR), abstraction (A), abstraction average rate (AVR), lifetime (LT), groundwater level (GWL), aquifer thickness (AT), depth from the surface to well screen (DSWS), distance from sea shoreline (DSSL)) and climate parameters (total rainfall (R), relative humidity (RH), minimum temperature (Tmin), maximum temperature (Tmax), average temperature (Tavg), average wind speed (W), minimum wind speed (Wmin), and maximum wind speed (Wmax)), in addition to initial chloride concentration (ICC), were considered as input variables. The output variable was the final chloride concentration (FCC). 17 ANN models were developed by varying the identified input parameters. Additionally, the coefficient of determination (R2) and root mean squared error (RMSE) were used to select the best predictive model. The results demonstrate that the ANN 5 model with the combinations of [ICC, RR, A, AVR, LT, GWL, DSWS, AT, DSSL, W] produced excellent estimation in predicting the value of final chloride concentration with reported values of 0.977 and 0.022 for R2 and RMSE respectively. The proposed approach illustrates how the ANN modeling technique can be used to identify the key variables required for the most significant parameters affecting chloride concentration.

Application of artificial neural network modeling techniques to signal strength computation

This paper presents development of artificial neural network (ANN) models to compute received signal strength (RSS) for four VHF (very high frequency) broadcast stations using measured atmospheric parameters. The network was trained using Levenberg-Marquardt back-propagation (LMBP) algorithm. Evaluation of different effects of activation functions at the hidden and output layers, variation of number of neurons in the hidden layer and the use of different types of data normalisation were systematically applied in the training process. The mean and variance of calculated MSE (mean square error) for ten different iterations were compared for each network. From the results, the ANN model performed reasonably well as computed signal strength values had a good fit with the measured values. The computed MSE were very low with values ranging between 0.0027 and 0.0043. The accuracy of the trained model was tested on different datasets and it yielded good results with MSE of 0.0069 for one dataset and 0.0040 for another dataset. The measured field strength was also compared with ANN and ITU-R P. 526 diffraction models and a strong correlation was found to exist between the measured field strength and ANN computed signals, but no correlation existed between the measured field strength and the predicted field strength from diffraction model. ANN has thus proved to be a useful tool in computing signal strength based on atmospheric parameters.

Gluten-free cookies with low glycemic index and glycemic load: optimization of the process variables via response surface methodology and artificial neural network

This research investigates the effect of baking temperature and time on the resistant starch (RS), glycemic index (GI) and glycemic load (GL) of gluten-free cookies, optimized the processing parameter using a chemometrics approach of response surface methodology (RSM) and artificial neural network (ANN). The in-vitro starch digestibility of the formulated cookies exhibited a monophasic starch digestogram. Increase in resistant starch, and a decrease in the predicted GI of the cookies, was associated with low temperature and high baking time. The use of RSM and ANN modelling techniques accurately predict the RS, pGI and GL (coefficient of determinant, R2 > 0.93 and root mean square of error = 0.43–0.62) of the gluten-free cookies. The optimal condition for the production of cookies with high RS, low pGI and GL were baking temperature of 158 °C and baking time of 20 min with predicted RS value of 19.61 g/100g of dry starch, pGI value of 56.98 and GL value 52.64.

Improving the resistant starch in succinate anhydride‐modified cardaba banana starch: A chemometrics approach

Resistant starch of succinate anhydride-modified cardaba banana starch was model using response surface methodology (RSM) and artificial neural network (ANN). This was investigated by varying the octenyl succinic anhydride (OSA) concentration (3%–5%), time (30–60 min), and pH (8–10) using a Box–Behnken design. According to the result, both the RSM and ANN modeling techniques predict the modification process with high accuracy (coefficient of determinant, R2 > .97 and root mean square of error =.44–.47). The optimization of the process parameters to maximize the resistant starch was performed using RSM and ANN coupled with genetic algorithm (GA). The ANN-GA gave the highest resistant starch content (39.87%) under the optimal process condition of OSA concentration 4.38%, reaction time 59.79 min, and 8.01 pH. This study revealed that the ANN-GA proved to be superior to RSM in the optimization of cardaba starch modification process. Practical applications In this study, the optimum starch modification conditions (succinate anhydride concentration, time, and pH) for maximizing the resistant starch in cardaba banana were investigated using response surface methodology and artificial neural network. The result obtained will enable the development of appropriate processes for starch modification from cardaba banana. It would then be possible to predict efficiently the conditions of succinate anhydride concentration, time, and pH, which will facilitate the maximization of resistant starch in the banana and increase its potential for use as a functional food ingredient in the food product development.

Comparative study of artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS) and multiple linear regression (MLR) for modeling of Cu (II) adsorption from aqueous solution using biochar derived from rambutan (Nephelium lappaceum) peel.

Presence of copper within water bodies deteriorates human health and degrades natural environment. This heavy metal in water is treated using a promising biochar derived from rambutan (Nephelium lappaceum) peel through slow pyrolysis. This research compares the efficacies of artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS), and multiple linear regression (MLR) models and evaluates their capability in estimating the adsorption efficiency of biochar for the removal of Cu (II) ions based on 480 experimental sets obtained in a laboratory batch study. The effects of operational parameters such as contact time, operating temperature, biochar dosage, and initial Cu (II) ion concentration on removing Cu (II) ions were investigated. Eleven different training algorithms in ANN and 8 different membership functions in ANFIS were compared statistically and evaluated in terms of estimation errors, which are root mean squared error (RMSE), mean absolute error (MAE), and accuracy. The effects of number of hidden neuron in ANN model and fuzzy set combination in ANFIS were studied. In this study, ANFIS model with Gaussian membership function and fuzzy set combination of [4 5 2 3] was found to be the best method, with accuracy of 90.24% and 87.06% for training and testing dataset, respectively. Contribution of this study is that ANN, ANFIS, and MLR modeling techniques were used for the first time to study the adsorption of Cu (II) ions from aqueous solutions using rambutan peel biochar.

Machine Learning for Pavement Performance Modelling in Warm Climate Regions

Accurate pavement performance modelling is an essential requirement for cost-effective pavement design and enhances pavement management decision making. Due to the complexity of the pavement structure and pavement response, dominant failure mechanisms may vary depending on the climate region: cold or warm. This study investigated the significance of pavement design factors on pavement performance in warm regions and compared them to set of factors previously identified for cold regions. An artificial neural network (ANN) supported by a forward sequential feature selection algorithm was employed to identify the most significant design factors prevailing in warm climate regions using data extracted from the Long-Term Pavement Performance database. In addition, five machine learning techniques were utilized to model the pavement performance in warm regions, namely: regression tree, support vector machine, ensembles, Gaussian process regression, and ANN. Moreover, conventional regression modelling was used for comparison assessment. The analysis revealed seven design factors that are significantly impacting asphalt pavement performance in warm regions: initial roughness, relative humidity, average wind velocity, average albedo, average emissivity, traffic volume, and pavement structural capacity. The results indicate that pavement performance in warm climate regions is dominated by different environmental factors than those found for cold climate regions. The ANN modelling technique produced the most accurate asphalt pavement performance models.

Artificial neural network modelling technique in predicting Western Australian seasonal rainfall

This paper presents the efficiency of non-linear modelling technique in predicting long-term seasonal rainfall of Western Australia. One of the commonly used non-linear modelling approaches, artificial neural network (ANN) was adopted for the construction of the non-linear models. The models were developed considering the past values of El Nino southern oscillation (ENSO) and Indian Ocean Dipole (IOD) as the probable influential variables of rainfall. The ANN models were constructed adopting the algorithm proposed by Lavenberg-Marquardt. The models were developed and tested for three rainfall stations in Western Australia. The models showed good generalisation capability of Western Australian spring rainfalls with Pearson correlations varying from 0.46 to 0.82 during the training phase and 0.55 to 0.96 during the testing phase. The errors and index of agreement of the IOD-ENSO based ANN models were also acceptable to be applied for rainfall forecasting.

Artificial neural network modelling technique in predicting Western Australian seasonal rainfall

Artificial neural network modelling technique in predicting Western Australian seasonal rainfall

Scavenging carbon deposition on alumina supported cobalt catalyst during renewable hydrogen-rich syngas production by methane dry reforming using artificial intelligence modeling technique

Methane dry reforming is a thermo-catalytical process that utilizes two principal components of greenhouse gases for the production of hydrogen-rich syngas . One major shortcoming of the methane dry reforming as a potential route for renewable hydrogen-rich syngas production is catalyst deactivation through carbon deposition . In this study, an artificial neural network approach was employed for predictive modeling of the deactivation of alumina supported cobalt catalyst used to catalyze methane dry reforming reaction. The effect of methane/carbon dioxide (CH 4 /CO 2 ) ratio, reaction temperature and nitrogen (N 2 ) flowrate on the carbon deposition were investigated using full factorial experimental design . Two artificial neural network modeling techniques namely multilayer perceptron neural network (MLPNN) and radial basis function (RBFNN) were employed for the prediction of carbon deposition per gram catalyst using data obtained from 170 experimental runs. The hidden neurons were optimized to obtain 16 and 20 units respectively for the MLPNN and the RBFNN resulting in the network architecture of 3, 16, 1 and 3, 20, 1, respectively. The statistical analysis of the network performance resulted in mean standard error (MSE) values of 0.048 and 0.00285 for training the MLPNN algorithm above and below stoichiometric conditions with corresponding R 2 values of 0.945 and 0.965. While MSE values of 0.0073 and 0.00015 were obtained for the training of the RBFNN algorithm above and below stoichiometric conditions with R 2 of 0.987 for both cases. Base on the statistical analysis the RBFNN model was adjudicated as a better predictor of the carbon deposition during the hydrogen-rich syngas production than the MLPNN model. The three input parameters were found to have varying levels of importance in the prediction of the carbon deposition. The reaction temperature was observed to be the most important parameters that influence the prediction of carbon deposition above stochiometric while CH 4 /CO 2 was the most important parameters that influence the prediction of carbon deposition below stoichiometric conditions. • MLPNN and RBFNN models were developed and employed to predict carbon deposition on Co/αAl 2 O 3 . • Optimized hidden neurons with 16 and 20 units respectively were obtained for MLPNN and RBFNN. • Levenberg-Marquardt algorithm was employed to train the MLPNN and RBFNN. • Both MLPNN and RBFNN accurately predict carbon deposition with minimal residuals. • RBFNN with MSE of 0.00015 and R 2 of 0.985 predicts better than MLPNN with MSE of 0.00285 and R 2 of 0.965