Accuracy Of Artificial Neural Network Research Articles

Variable selection and training set design for particle classification using a linear and a non-linear classifier

While particulate products are often characterized by their median diameter or the width of the particle size distribution, information is rarely given about the agglomeration degree of the product. To obtain this information, a tool combining image analysis and discriminant factorial analysis (DFA) was introduced in previous works. The accuracy of that method depended on the number of image descriptors selected, i.e. measurements describing each particle: few image descriptors resulted in rather poor classification while too many lead to an overfitting of the data. The aim of this study is twofold: First, we want to compare the classification accuracy of artificial neural networks (ANN) and DFA which, contrary to ANN, forms linear classifiers. Second, we want to provide an easy-to-implement procedure for generating particle classifiers. We used a qualitative measure called Proportional Similarity to test whether a subset selection of image descriptors was necessary to avoid an overfitting. The influence of the training set size was investigated as well as the transferability of the classifier on data obtained under different experimental conditions. The chemical systems used were l-alanine/water and adipic acid/water and the classes considered were single crystals, agglomerates, and gas bubbles. The results show that an ANN classifier provides higher accuracy and is more effective when only few image descriptors are available while DFA is simpler to create. Moreover, we show good transferability of classifiers trained on data of different experimental conditions. Based on our results, we provide guidelines for classification of particulate systems.

Can artificial neural network and response surface methodology reliably predict hydrogen production and COD removal in an UASB bioreactor?

Artificial Neural Network (ANN) and Response Surface Methodology (RSM) modeling was employed in an Upflow Anaerobic Sludge Blanket (UASB) bioreactor for optimization of hydrogen yield and COD (Chemical Oxygen Demand) removal efficiency. Experimental data were generated by running seventeen fermentation experiments at varying hydraulic retention times, immobilized cell volumes and temperatures. RSM and ANN models predicted similar optimum conditions for these process parameters. Upon validation, the prediction error for ANN and RSM was observed to be 2.22 and 9.64% on hydrogen yield and 1.01 and 6.34% on COD removal. These results suggested a greater accuracy and higher reliability of ANN in modeling and optimizing the bioprocess parameter interactions associated to the fermentation process. In addition, the study demonstrated a higher molar biohydrogen yield (0.90 mol-H2/mol glucose) and COD removal efficiency (84.81%) in the UASB system optimized by ANN modeling.

A Hybrid Approach Based on Genetic Algorithm and Particle Swarm Optimization to Improve Neural Network Classification

Artificial Neural Network (ANN) has played a significant role in many areas because of its ability to solve many complex problems that mathematical methods failed to solve. However, it has some shortcomings that lead it to stop working in some cases or decrease the result accuracy. In this research the authors propose a new approach combining particle swarm optimization algorithm (PSO) and genetic algorithm (GA), to increase the classification accuracy of ANN. The proposed approach utilizes the advantages of both PSO and GA to overcome the local minima problem of ANN, which prevents ANN from improving the classification accuracy. The algorithms start with using backpropagation algorithm, then it keeps repeating applying GA followed by PSO until the optimum classification is reached. The proposed approach is domain independent and has been evaluated by applying it using nine datasets with various domains and characteristics. A comparative study has been performed between the authors' proposed approach and other previous approaches, the results show the superiority of our approach.

Enhanced Monthly Precipitation Forecasting Using Artificial Neural Network and Singular Spectrum Analysis Conjunction Models

Precipitation is a complex atmospheric process with both temporal and spatial variations which make it difficult to forecast. Several models have been developed to forecast precipitation. This study investigates the accuracy of artificial neural network (ANN) and singular spectrum analysis (SSA) conjunction model in monthly precipitation forecasting. Here, SSA is used to preprocess raw input signals to provide data of high quality to ANN. The conjunction model is obtained by combining two methods, ANN and SSA, and compared with the single/regular ANN. In SSA, the window length and the optimum number of components must be determined. For this purpose, several tentative models, one for each of the available component numbers, are developed, and the related component number of the model with best performance is considered as the optimum number of components. If the selected optimum number is less or equal to the chosen window length which in turn indicates the adequacy of the chosen window length. Monthly precipitation data from one rain gauge station, Ponel station, in northern Iran, is used in the study. The root mean square error (RMSE), correlation coefficient (R) and coefficient of efficiency (CE) statistics are used as the comparing criteria. The comparison of the results reveals that the conjunction models could increase the forecast accuracy of the ANN model in monthly precipitation forecasting. It is found that the conjunction model with $$RMSE = 52.257$$ , $$R = 0.858$$ and $$CE = 0.731$$ in the validation/testing period is superior in forecasting monthly precipitations than the most accurate ANN model with $$RMSE = 91.096$$ , $$R = 0.444$$ and $$CE = 0.183$$ , respectively.

Ultrasonic treatment of water contaminated with various pollutants onto copper nanowires loaded on activated carbon using response surface methodology and artificial intelligent

In this study, the potential application of copper nanowires loaded on activated carbon for simultaneous removal of Disulfine blue (DB), Crystal violet (CV) and Sunset yellow (SY) has been described. The relation between adsorption properties with variables such as solution pH, adsorbent value, contact time and initial dyes concentration was investigated and optimized. A three‐layer artificial neural network (ANN) model was utilized to predict dyes removal (%) by adsorbent following conduction of experiments. The training of network at above mention experimental data confirms its ability to forecast the removal performance with a linear transfer function (purelin) at output layer. The Levenberg–Marquardt algorithm and tangent sigmoid transfer function (tansig) with 16 neurons at the hidden layer was applied. Parameters were optimized by central composite design (CCD) combined with response surface methodology (RSM) and desirability function. The accuracy of ANN was judged according to both MSE and AAD% at optimal conditions and results indicate its superiority to RSM model in term of higher R2 and lower AAD% values. This observation was also corroborated by the parity plots between the predicted and experimental values. The ANN model was better in both data fitting and prediction capability in comparison to RSM model.

Experimental evaluation, sensitivity analyzation and ANN modeling of thermal conductivity of ZnO-MWCNT/EG-water hybrid nanofluid for engineering applications

The thermal conductivity of ZnO-MWCNT / EG-water hybrid nanofluid was experimentally investigated at solid volume fractions of 0.02–1% and temperatures of 30–50°C. The thermal conductivity of a nanofluid was measured in the 50%:50% mixture of ethylene glycol (EG) and water. The results showed that the maximum thermal conductivity ratio (TCR) is obtained at 50 ˚C and at the solid volume fraction of 1%. A new correlation was proposed for prediction of experimental TCR. The coefficient of determination (R-squared) of this correlation and its maximum regression error were equal to 0.9864 and 2.6% respectively. The sensitivity measurement results showed that the hybrid nanofluid shows the highest sensitivity at the highest temperature and solid volume fraction. An artificial neural network (ANN) was developed based on the experimental TCR data. In the best case, an ANN with two hidden layers and four neurons in each layer was obtained. Comparisons of experimental TCR with ANN and correlation output indicated high capability and accuracy of ANN in modeling of TCR data.

Empirical Correlations and an Artificial Neural Network Approach to Estimate Saturated Vapor Pressure of Refrigerants

The examination of available vapor pressure data in the case of the methane, ethane, propane and butane halogenated refrigerants, allowed recommendations of standard equations for this property. In this study, three new models include a general correlation; a substance-dependent correlation and an artificial neural network (ANN) approach have been developed to estimate the saturated vapor pressure of refrigerants. With the presented approaches, vapor pressures have been calculated and compared with source data bank for 5600 data points of 28 refrigerants. The accuracies of new correlations and ANN have been compared with most commonly used correlations and the comparison indicates that all new models provide more accurate results than other literature correlations considered in this work.

DIVERSIDADE FLORÍSTICA E EQUABILIDADE EM FRAGMENTOS FLORESTAIS USANDO REDES NEURAIS ARTIFICIAIS

Este estudo teve como objetivo avaliar a eficiência da predição dos índices de diversidade de Shannon (H’) e de Equabilidade de Pielou (J) em fragmentos florestais do Cerrado brasileiro a partir de índices de vegetação e métricas da paisagem empregando redes neurais artificiais (RNA). Utilizaram-se redes anteroalimentadas (feedforward), treinadas por meio do algoritmo da retropropagação do erro (back propagation). As variáveis utilizadas como entradas das RNA para a estimação simultânea dos índices foram: as categóricas (índices H’ e J) e as numéricas relacionadas às médias e desvios padrão dos índices de vegetação (NDVI, SAVI, EVI, MVI5 e MVI7) e métricas da paisagem (AREA, GYRATE, SHAPE, CONTIG, CORE e ENN). Foram gerados cinco modelos de RNA a partir das relações funcionais entre as variáveis numéricas inerentes aos índices de vegetação em duas épocas, uma seca (junho) e outra chuvosa (fevereiro). A arquitetura das redes foi a Multilayer Perceptron (MLP) para estimar simultaneamente H’ e J: 500 utilizando os índices de vegetação na época úmida (100 para cada índice de vegetação) e 500, na seca (100 para cada índice de vegetação). Foi avaliada a precisão, acurácia e realismo biológico das RNA. As redes construídas na época chuvosa e seca que utilizaram os índices de vegetação MVI5 (Moisture Vegetation Index) e SAVI (Soil Adjusted Vegetation Index), respectivamente, foram mais adequadas, precisas e realistas biologicamente para estimar, simultaneamente, os índices de H’ e de J. A modelagem por RNA demonstrou-se adequada para estimar os índices de diversidade e equabilidade.

Comparing six pixel-wise classifiers for tropical rural land cover mapping using four forms of fully polarimetric SAR data

ABSTRACTThis study evaluates four commonly used forms of synthetic aperture radar (SAR) data for land-cover classification in tropical rural areas. The backscatter coefficient of linearly polarized L-band SAR was compared to two distinctive feature sets derived from Eigen-based and model-based decompositions. The performance of six classifiers available in Orfeo Toolbox (OTB), that is, Bayes, artificial neural networks (ANNs), Support Vector Machine (SVM), decision trees, Random Forests (RFs), and gradient boosting trees (GBTs), was investigated to distinguish five and seven land-cover classes, with particular attention given to several types of woody vegetation: forest, mixed garden, rubber, oil palm, and tea plantations. Classifiers reacted differently to ingested forms of SAR data, and careless use of data input yielded a negative impact. The results showed that SVM provided the highest overall accuracy although the performance was not significantly better than the others. Tuning the parameters, however, significantly improved the accuracy of ANN and SVM, while RF and GBT did not respond well. Responses of two SVM parameters (cost and kernel type) fluctuated somewhat, which required further attention. ANN accuracy was improved when the number of neurons in the hidden layer was set between 10 and 12. We found that accuracy imbalance existed between designated land-cover classes, especially in woody vegetation. Imbalance can partially be reduced by tuning specific classifiers. We showed that classifier tuning can lead to significantly improved accuracy, especially for classes having medium or low accuracies. This research also demonstrated that freely available toolkits such as OTB and QGIS can be beneficial for mapping activities in developing countries, achieving a reasonable accuracy if the classification parameters are tuned properly.

Prediction and Experimental Study on Cutting Force of Austempered Vermicular Graphite Cast Iron Using Artificial Neural Network

In this study, a technique is proposed to predict cutting force of austempered vermicular graphite cast irons (VGCI) which are widely used in the industry by using neural network. The effect of austempering heat treatment on the cutting force was experimentally achieved. The samples were austenitized at 900 °C for 90 minutes and then austempered at different temperatures (320 °C and 370 °C) for 60, 90 and 120 minutes. Machinability tests were carried on under dry conditions at the CNC machining center with the cutting parameters selected in accordance with ISO 3685. In the experiment, cutting force depending on hardness, cutting speed and feed rate were measured. These results were used for input parameters (training, testing and validation) of Artificial Neural Network (ANN) and prediction model was developed. Output value of ANN and experimental results were compared and accuracy of ANN is 99.99% and 99.62% for training and test values, respectively.DOI: http://dx.doi.org/10.5755/j01.mech.23.1.13699

Data-driven nonlinear modeling studies on removal of Acid Yellow 59 using Si-doped multi-walled carbon nanotubes

Data-driven modeling of removal of color index name of Acid Yellow 59 from aqueous solutions using multi-walled carbon nanotubes by multiple (non)linear regression and artificial neural networks (ANN) models based on leave-one-out cross-validation to predict the adsorbed dye amount per unit mass of adsorbent (mg g−1) and performance evaluation of the proposed multiple (non)linear regression and ANN models is the main novel contributor of the present study. Initial dye concentration, adsorbent concentration, reaction time, and temperature were determined as explanatory variables and input neurons for multiple (non)linear regression and ANN models, respectively. The total number of experiments was determined as 1280 statistically. The results showed that multilayer perception ANN model ( $$R^{2}_{\text{training}}$$ = 0.9997, $$R^{2}_{\text{testing}}$$ = 0.9993, RMSE = 0.7678, MAE of 0.0007) predicted q t better than multiple (non)linear regression model ( $$R^{2}_{\text{adj}}$$ = 0.9645, $$R^{2}_{\text{pred}}$$ = 0.9633, SE = 9.55) and MLR (R 2 = 0.9543, SE = 10.87) models. The results justified the accuracy of ANN in prediction of q t , significantly.

Using a Feature Subset Selection method and Support Vector Machine to address curse of dimensionality and redundancy in Hyperion hyperspectral data classification

The curse of dimensionality resulted from insufficient training samples and redundancy is considered as an important problem in the supervised classification of hyperspectral data. This problem can be handled by Feature Subset Selection (FSS) methods and Support Vector Machine (SVM). The FSS methods can manage the redundancy by removing redundant spectral bands. Moreover, kernel based methods, especially SVM have a high ability to classify limited-sample data sets. This paper mainly aims to assess the capability of a FSS method and the SVM in curse of dimensional circumstances and to compare results with the Artificial Neural Network (ANN), when they are used to classify alteration zones of the Hyperion hyperspectral image acquired from the greatest Iranian porphyry copper complex. The results demonstrated that by decreasing training samples, the accuracy of SVM was just decreased 1.8% while the accuracy of ANN was highly reduced i.e. 14.01%. In addition, a hybrid FSS was applied to reduce the dimension of Hyperion. Accordingly, among the 165 useable spectral bands of Hyperion, 18 bands were only selected as the most important and informative bands. Although this dimensionality reduction could not intensively improve the performance of SVM, ANN revealed a significant improvement in the computational time and a slightly enhancement in the average accuracy. Therefore, SVM as a low-sensitive method respect to the size of training data set and feature space can be applied to classify the curse of dimensional problems. Also, the FSS methods can improve the performance of non-kernel based classifiers by eliminating redundant features.

Determining the importance of soil properties for clay dispersibility using artificial neural network and daptive neuro-fuzzy inference system

The main purpose of the current research is comparing the results of Artificial Neural Network (ANN) with Adaptive Neuro-Fuzzy Inference System (ANFIS) with regard to determination of the importance of soil properties affecting clay dispersibility. After taking samples from two depths of 0-40 and 40-80 cm, the spontaneous and mechanical dispersions of clay were recorded using both weighing and turbidimetric methods. To determine the degree of importance of soil properties affecting clay dispersibility, first ANNs and ANFIS in MATLAB Software were determined, using all research variables. After determining less effective properties and omitting them, the mentioned networks with the remaining variables including percentage of clay and sand, soil reaction, Electrical Conductivity (EC) and Sodium Adsorption Ratio (SAR) were measured and the degree of importance of each variable in clay dispersibility was determined. Finally, the results of ANNs and ANFIS were compared by calculation of validation parameters. Existence of high correlation between calculated values for weighing and turbidimetric methods showed a linear relationship between the two methods. In general, in both depths and for both weighing and turbidimetric methods, the sensitivity of clay dispersibility to the percentage of the clay, sand and SAR, was higher than any other variable. Although the results obtained from the validation statistics indicate high accuracy of both ANN and ANFIS models, the last model showed relatively better results as compared to ANN model.

Artificial neural network and regression coupled genetic algorithm to optimize parameters for enhanced xylitol production by Debaryomyces nepalensis in bioreactor

In this study, prediction ability and optimization ability of regression and artificial neural network (ANN) models were compared. The input variables used to predict xylose consumption, biomass and xylitol production by regression analysis and neural network model are temperature, fermentation time, pH, kLa, biomass and glycerol of previous data points. Determination of coefficient (R2) was used to assess the adequacy of the regression model and R2 for xylitol and biomass were 86.56% and 96.43% respectively. A multi layered feed forward neural network (ANN) of 5-10-2 topology has been developed to predict the xylitol production. Results showed that prediction accuracy of ANN was apparently higher when compared to regression model. Genetic algorithm (GA) was used to find the optimum parameters to enhance xylitol production. Optimization of fermentation parameters was carried out using hybrid regression GA and hybrid ANN GA with an optimum prediction error of 10% and 3.5% respectively. ANN coupled GA is considered to be the better optimization method due to its high accuracy and low prediction error and recommended to be employed for optimization of fermentative process.

Relative Importance of Intellectual Capital Determinants, Using an Artificial Neural Network Approach: Case Study - Mellat Bank of Iran

Intellectual capital (IC) is the main part of intangible assets and provides competitive advantages for organizations. Therefore, evaluating these assets has a significant value for companies. The aim of this study is to determine the relative importance of IC determinants. Two kinds of questionnaires related to IC were used to collect data for Artificial Neural Network (ANN) and Analytic Hierarchy Process (AHP) methods. In the next step the accuracy of ANN is compared with Multiple Linear Regression and Nonlinear Least Square Fitting methods. Factor analysis is also applied to validate the model and reduction of indicators. Finally, to gain assurance about the ANN results, the results are compared with the AHP’s outcome. The obtained importance order of three IC determinants in the studied banking system is relational capital, human capital and structural capital respectively.

Evaluating the accuracy of ANN and LR models to estimate the water quality in Zarivar International Wetland, Iran

One of the most important qualitative aspects of wetland ecosystem management is preserving the natural quality of water in such environments. This would not be achievable unless continuous water quality monitoring is implemented. With the recent advances in remote sensing technology, this technology could assist us to produce accurate models for estimating water quality variables in the ecosystem of wetlands. The present study was carried out to evaluate the capability of remote sensing data to estimate the water quality variables [pH, total suspended solids (TSS), total dissolved solids (TDS), turbidity, nitrate, sulfate, phosphate, chloride and the concentration of chlorophyll a] in Zarivar International Wetland using linear regression (LR) and artificial neural network (ANN) models. For this purpose, spectral reflectance of bands 2, 3, 4 and 5 of the OLI sensor of Landsat 8 was utilized as the input data and the collected chemical and physical data of water samples were selected as the objective data for both ANN and LR models. Based on our results overall, ANN model was the proper model compared with LR model. The spectral reflectance in bands 5 and 4 of OLI sensor revealed the best results to estimate TDS, TSS, turbidity and chlorophyll in comparison with other used bands in ANN model, respectively. We conclude that OLI sensor data are an excellent means for studying physical properties of water quality and comparing its chemical properties.

Modeling and simulation of xylitol production in bioreactor by Debaryomyces nepalensis NCYC 3413 using unstructured and artificial neural network models

This study examines the use of unstructured kinetic model and artificial neural networks as predictive tools for xylitol production by Debaryomyces nepalensis NCYC 3413 in bioreactor. An unstructured kinetic model was proposed in order to assess the influence of pH (4, 5 and 6), temperature (25°C, 30°C and 35°C) and volumetric oxygen transfer coefficient kLa (0.14h−1, 0.28h−1 and 0.56h−1) on growth and xylitol production. A feed-forward back-propagation artificial neural network (ANN) has been developed to investigate the effect of process condition on xylitol production. ANN configuration of 6-10-3 layers was selected and trained with 339 experimental data points from bioreactor studies. Results showed that simulation and prediction accuracy of ANN was apparently higher when compared to unstructured mechanistic model under varying operational conditions. ANN was found to be an efficient data-driven tool to predict the optimal harvest time in xylitol production.

Fault detection of engine timing belt based on vibration signals using data-mining techniques and a novel data fusion procedure

In this research, an intelligent procedure was designed and implemented based on vibration signals for detecting and classifying prevalent faults of an internal combustion engine timing belt. The vibration signals of the timing belt were captured during operation in six different states: healthy, tooth crack, back crack, wear, separated tooth, and oil pollution. These signals were processed at three domains, namely, time, frequency, and time–frequency domains. Time-domain signals were transformed into the frequency and time–frequency domains using fast Fourier transform and wavelet transform, respectively. Then, six statistical features were extracted from vibration signals at all three domains. The extracted features were used as inputs to an artificial neural network for the primary classification of timing belt defects. Classification accuracy of artificial neural network in detecting and classifying timing belt faults in the time, frequency, and time–frequency domains have obtained 71%, 78%, and 84%, respectively. Combining separate classification accuracies from time, frequency, and time–frequency domains has been implemented using Dempster–Shafer theory of evidence. Classification accuracy based on the fusion of time- and frequency-domain classifiers was 97%, from time and time–frequency results was 98%, and from frequency and time–frequency results was also 98%, whereas the combination of results for all domains led to a >99% accuracy. Results show that the proposed methodology can detect and classify timing belt defects with high precision and reliability before failure occurrence.

Improvement in artificial neural network-based estimation of grid connected photovoltaic power output

This paper presents a method to improve the accuracy of artificial neural network (ANN)–based estimation of photovoltaic (PV) power output by introducing two more inputs, solar zenith angle and solar azimuth angle, in addition to the most widely used environmental information, plane-of-array irradiance and module temperature. Solar zenith angle and solar azimuth angle define the solar position in the sky; hence, the loss of modeling accuracy due to impacts of solar angle-of-incidence and solar spectrum is reduced or eliminated. The observed data from two sites where local climates are significantly different is used to train and test the proposed network. The good performance of the proposed network is verified by comparing with existing ANN model, algebraic model, and polynomial regression model which use environmental information only of plane-of-array irradiance and module temperature. Our results show that the proposed ANN model greatly improves the accuracy of estimation in the long term under various weather conditions. It is also demonstrated that the improvement in estimating outdoor PV power output by adding solar zenith angle and azimuth angle as inputs is useful for other data-driven methods like support vector machine regression and Gaussian process regression.

A computational approach for prediction of donor splice sites with improved accuracy

Identification of splice sites is important due to their key role in predicting the exon-intron structure of protein coding genes. Though several approaches have been developed for the prediction of splice sites, further improvement in the prediction accuracy will help predict gene structure more accurately. This paper presents a computational approach for prediction of donor splice sites with higher accuracy. In this approach, true and false splice sites were first encoded into numeric vectors and then used as input in artificial neural network (ANN), support vector machine (SVM) and random forest (RF) for prediction. ANN and SVM were found to perform equally and better than RF, while tested on HS3D and NN269 datasets. Further, the performance of ANN, SVM and RF were analyzed by using an independent test set of 50 genes and found that the prediction accuracy of ANN was higher than that of SVM and RF. All the predictors achieved higher accuracy while compared with the existing methods like NNsplice, MEM, MDD, WMM, MM1, FSPLICE, GeneID and ASSP, using the independent test set. We have also developed an online prediction server (PreDOSS) available at http://cabgrid.res.in:8080/predoss, for prediction of donor splice sites using the proposed approach.