Linear Artificial Neural Network Research Articles

Structure–activity relationship (SAR) modelling of mosquito larvicides

An attempt was made to derive structure–activity models allowing the prediction of the larvicidal activity of structurally diverse chemicals against mosquitoes. A database of 188 chemicals with their activity on Aedes aegypti larvae was constituted from analysis of original publications. The activity values were expressed in log 1/IC50 (concentration required to produce 50% inhibition of larval development, mmol). All the chemicals were encoded by means of CODESSA and autocorrelation descriptors. Partial least squares analysis, classification and regression tree, random forest and boosting regression tree analyses, Kohonen self-organizing maps, linear artificial neural networks, three-layer perceptrons, radial basis function artificial neural networks and support vector machines with linear, polynomial, radial basis function and sigmoid kernels were tested as statistical tools. Because quantitative models did not give good results, a two-class model was designed. The three-layer perceptron significantly outperformed the other statistical approaches regardless of the threshold value used to split the data into active and inactive compounds. The most interesting configuration included eight autocorrelation descriptors as input neurons and four neurons in the hidden layer. This led to more than 96% of good predictions on both the training set and external test set of 88 and 100 chemicals, respectively. From the overall simulation results, new candidate molecules were proposed which will be shortly synthesized and tested.

Quantitative assessment of sewer overflow performance with climate change in northwest England

Changes in rainfall patterns associated with climate change can affect the operation of a combined sewer system, with the potential increase in rainfall amount. This could lead to excessive spill frequencies and could also introduce hazardous substances into the receiving waters, which, in turn, would have an impact on the quality of shellfish and bathing waters. This paper quantifies the spilling volume, duration and frequency of 19 combined sewer overflows (CSOs) to receiving waters under two climate change scenarios, the high (A1FI), and the low emissions (B1) scenarios, simulated by three global climate models (GCMs), for a study catchment in northwest England. The future rainfall is downscaled, using climatic variables from HadCM3, CSIRO and CGCM2 GCMs, with the use of a hybrid generalized linear–artificial neural network model. The results from the model simulation for the future in 2080 showed an annual increase of 37% in total spill volume, 32% in total spill duration, and 12% in spill frequency for the shellfish water limiting requirements. These results were obtained, under the high emissions scenario, as projected by the HadCM3 as maximum. Nevertheless, the catchment drainage system is projected to cope with the future conditions in 2080 by all three GCMs. The results also indicate that under scenario B1, a significant drop was projected by CSIRO, which in the worst case could reach up to 50% in spill volume, 39% in spill duration and 25% in spill frequency. The results further show that, during the bathing season, a substantial drop is expected in the CSO spill drivers, as predicted by all GCMs under both scenarios.Editor Z.W. Kundzewicz; Associate editor L. See

NMR Log Prediction from Seismic Attributes: Using Multiple Linear Regression and Neural Network Methods

This study proposes a strategy to make a quantitative correlation between the NMR log-derived free fluid porosity and seismic attributes using multiple linear regression and artificial neural network. At first the well logs tied to seismic data by creating a synthetic seismogram at each well using the sonic and density logs. Then the post-stack seismic data will be inverted to acoustic impedance by applying the created synthetic seismograms. The higher resolution of well logs than seismic data problem is dissolved by three-dimensional modeling and averaging all of the seismic and logs data in each model cell. Then the cell’s properties are processed as inputs and outputs for all operations. Stepwise regression is used to select the best attributes predicting the free fluid porosity. The multiple linear regression equation and its correlation at each step are shown. Finally, an artificial neural network of multi-layer neural network type with back propagation of errors is used for the prediction, and the better method between these two methods was selected to apply it to the whole volume of study. The proposed methodology was applied to the South Pars Gas Field in the Persian Gulf Basin. The seismic attributes were extracted from three separated seismic parts. The petrophysical logs from four wells in these seismic cubes were used for generating and evaluating the reliability of the neural network.

Work Piece Identification based on Plasma Emission Analysis for Customized Laser Processing

A study on the identification of the work piece being laser machined is conducted with the goal of developing a strategy to automatically adjust the processing parameters. The identification technique is based on the spectroscopic analysis of the plasma emission generated during laser ablation. Radiation emitted by the plasma is characteristic of the ablated material and thus can be used as a spectral “fingerprint” for identification purposes. Linear correlation and artificial neural networks are implemented as classification algorithms. Both methods present efficient identification of samples with recognizably different emission spectra. Neural networks outperform linear correlation for the identification of work pieces with very similar spectra. Additionally, the influence of processing and acquisition conditions on the performance of both algorithms is investigated. Finally, a strategy for customized processing based on the identification of each studied work piece is proposed. Linear correlation is used for general identification. Meanwhile, artificial neural networks are solely employed to identify materials for which classification performance of linear correlation decreases. Subsequently, machining parameters are adjusted according to the identified work piece.

Prediction of Thermophysical Properties of Helium Using Linear Prediction and Artificial Neural Networks

Thermophysical properties of helium are significant in practical applications. However, the values of properties vary under different circumstances, which may have bad impacts on practical productions and applications. In our study, computational models like Linear Prediction and Artificial Neural Networks (ANNs) are applied to predict the thermophysical properties of the chemical substances. By analyzing 50 data groups using Linear Prediction, General Regression Neural Network (GRNN) and Multilayer Feedforward Neural Network (MLFN) methods, 9 models were successfully established to predict the thermophysical properties of helium, including density, energy, enthalpy, entropy, isochoric heat capacity, isobaric heat capacity, viscosity, thermal conductivity and dielectric constant. Within permissible error range (30% tolerance), our models were proved to be robust and accurate which indicates that ANN models can be used to predict the thermophysical properties of helium.

An empirical analysis of thermal protective performance of fabrics used in protective clothing.

Fabric-based protective clothing is widely used for occupational safety of firefighters/industrial workers. The aim of this paper is to study thermal protective performance provided by fabric systems and to propose an effective model for predicting the thermal protective performance under various thermal exposures. Different fabric systems that are commonly used to manufacture thermal protective clothing were selected. Laboratory simulations of the various thermal exposures were created to evaluate the protective performance of the selected fabric systems in terms of time required to generate second-degree burns. Through the characterization of selected fabric systems in a particular thermal exposure, various factors affecting the performances were statistically analyzed. The key factors for a particular thermal exposure were recognized based on the t-test analysis. Using these key factors, the performance predictive multiple linear regression and artificial neural network (ANN) models were developed and compared. The identified best-fit ANN models provide a basic tool to study thermal protective performance of a fabric.

Comparison Of Linear And Non-Linear Calibration Models For Non-Destructive Firmness Determining Of ‘Mazafati’ Date Fruit By Near Infrared Spectroscopy

The selection of calibration method is one of the most important factors affecting the measurement accuracy with near infrared spectroscopy. In this research, the performance of two general calibration methods, namely, linear partial least squares regression and nonlinear back propagation artificial neural networks for firmness predicting of ‘Mazafati’ date fruit was investigated. A total of 175 date samples harvested during fruit ripening were selected as the data set. Optical scanning was performed with a fiber optic near infrared spectrometer with a range of 900–1700 nm. The inputs of back propagation artificial neural networks were the first five principal components resulting from the principal component analysis, namely, principal component analysis artificial neural network modeling or the first six latent variables obtained from partial least squares regression, namely, latent variable artificial neural network modeling. Both the leave-one-out cross validation and test set validation showed the priority of latent variable artificial neural network model with respect to partial least squares and also principal component analysis artificial neural network models. The best latent variable artificial neural network model could predict the firmness of Mazafati date fruits with Rp 2 of 0.90, RMSEP of 1.30 N, and SDRp of 3.28. The results also recommended the adoption of nonlinear latent variable artificial neural network modeling for applying light scattering properties of fruits in order to predict indirect property of firmness with near infrared spectroscopy.

Atmospheric Pollution: Statistical Models of Predicting SO<sub>2</sub> Concentration in Shanghai

Air quality has been deteriorated seriously in Shanghai as a result of urbanization and modernization. Meteorological conditions affect air pollution levels in the urban atmosphere significantly due to their important role in transport and dilution of the pollutants. This paper aims to investigate usability of statistical methods for prediction of SO2concentration. Multiple linear regression (MLR) model and artificial neural network (ANN) model were used to predict next days SO2concentrations. The calculatedR2values were 0.56 and 0.55 for MLR and ANN model, respectively. This result shows the MLR analysis, providing simple and faster application facilities, may also be useful for modeling the impacts of meteorological factors on SO2levels besides the time consuming ANN model.

Comparison of Classifiers for Decoding Sensory and Cognitive Information from Prefrontal Neuronal Populations

Decoding neuronal information is important in neuroscience, both as a basic means to understand how neuronal activity is related to cerebral function and as a processing stage in driving neuroprosthetic effectors. Here, we compare the readout performance of six commonly used classifiers at decoding two different variables encoded by the spiking activity of the non-human primate frontal eye fields (FEF): the spatial position of a visual cue, and the instructed orientation of the animal's attention. While the first variable is exogenously driven by the environment, the second variable corresponds to the interpretation of the instruction conveyed by the cue; it is endogenously driven and corresponds to the output of internal cognitive operations performed on the visual attributes of the cue. These two variables were decoded using either a regularized optimal linear estimator in its explicit formulation, an optimal linear artificial neural network estimator, a non-linear artificial neural network estimator, a non-linear naïve Bayesian estimator, a non-linear Reservoir recurrent network classifier or a non-linear Support Vector Machine classifier. Our results suggest that endogenous information such as the orientation of attention can be decoded from the FEF with the same accuracy as exogenous visual information. All classifiers did not behave equally in the face of population size and heterogeneity, the available training and testing trials, the subject's behavior and the temporal structure of the variable of interest. In most situations, the regularized optimal linear estimator and the non-linear Support Vector Machine classifiers outperformed the other tested decoders.

Optimization of Telephone Handset Distortion in Speaker Recognition by Discriminative Feature Design

This technique used mel-cepstral features, log spectrum and prosody based features with a non- linear artificial neural network in designing speaker recognition features that minimize telephone handset distortion. Effect of handset distortion was done to maximize speaker recognition performance specifically in the setting of telephone handset mismatch between training and testing as results on the 1998 NIST Speaker Recognition Evaluation corpus shows a high efficiency of 98% performance.

A Preliminary Study on Multivariate Prediction of Seizure Outcome after Epilepsy Surgery

Surgical outcomes of epilepsy surgery vary across patients, and clinicians need to estimate possible outcomes before surgery. The aim of this study was to identify predictors of seizure outcome one year after surgery for patients with drug-resistant epilepsy. Twenty-three patients with Temporal Lobe Epilepsy (TLE) who underwent surgery were included in the study. Their demographical information, seizure history, findings of EEG and neuroimaging tests (mainly Magnetic Resonance Imaging (MRI) and Magnetic Resonance Spectroscopy (MRS), intracranial EEG (icEEG) findings, seizure outcome and pathological findings were reviewed. Bivariate analyses were performed to examine the univariate association of each variable with the outcome, and exclude the most insignificant ones. The remaining data were randomly assigned to the training and test sets, and three multivariate analysis approaches (Logistic Regression (LR), Linear Discriminant Analysis (LDA) and Artificial Neural Network (ANN)) were performed repetitively. Model performance was compared using Receiver-Operating Characteristic (ROC) analysis. Resampling the data to the training and test sets resulted in large variations in the classification accuracies of each multivariate approach. The ROC results indicated that the medium classification performances were moderate. Important outcome predictors identified included EEG lateralization score, icEEG lateralization score, and the presence of Hippocampal Sclerosis (HS). The results suggested that multivariate models could predict seizure outcome after TLE surgery with moderate accuracy. Further studies are needed to improve prediction accuracy and identify reliable predictors of seizure outcome.

Chemometric estimation of post-mortem interval based on Na+ and K+ concentrations from human vitreous humour by linear least squares and artificial neural networks modelling

The subject of this paper is to determine the post-mortem interval (PMI) using the data obtained by potentiometric measurements of the electrolyte concentrations (potassium and sodium) in human vitreous humour. The data were processed by linear least squares (LLS) and artificial neural network (ANN) procedures. The LLS mathematical models have been developed as calibration models for prediction of the PMI. The quality of the models was validated by the leave one out (LOO) technique and by using an external data set. High agreement between experimental and predicted PMI values indicated the good quality of the derived models. Additionally, we analysed the influence of various factors (the cause of death, sex, differences between electrolyte concentrations in left and right eye) on the accuracy and reliability of obtained PMI. The ANN method was based on 174 forensic cases with different causes of death and known PMI ranging from 3.1–24.1 hours. The external data sets corresponding to 40 selected forensic cases were tested. Excellent correlation between experimental PMI and PMI predicted by ANN was obtained with a coefficient of correlation r2=0.9611. In comparison to the LLS regression method applied on the complete available data, the prediction of PMI with ANN was improved by1.66 hours.

Hybrid calibration method for six-component force/torque transducers of wind tunnel balance based on support vector machines

A hybrid calibration approach based on support vector machines (SVM) is proposed to characterize nonlinear cross coupling of multi-dimensional transducer. It is difficult to identify these unknown nonlinearities and crosstalk just with a single conventional calibration approach. In this paper, a hybrid model comprising calibration matrix and SVM model for calibrating linearity and nonlinearity respectively is built up. The calibration matrix is determined by linear artificial neural network (ANN), and the SVM is used to compensate for the nonlinear cross coupling among each dimension. A simulation of the calibration of a multi-dimensional sensor is conducted by the SVM hybrid calibration method, which is then utilized to calibrate a six-component force/torque transducer of wind tunnel balance. From the calibrating results, it can be indicated that the SVM hybrid calibration method has improved the calibration accuracy significantly without increasing data samples, compared with calibration matrix. Moreover, with the calibration matrix, the hybrid model can provide a basis for the design of transducers.

Application of artificial neural networks to determine the authentication of fattening diets of Iberian pigs according to their triacylglycerol profiles

The triacylglycerols in the subcutaneous fat from Iberian pigs reared on four different feeding types, Montanera, Recebo, extensive Cebo and intensive Cebo, have been determined by gas chromatography with a flame ionization detector. Analyses were performed in a column coated with a bonded stationary phase (50% phenyl-50% methylpolysiloxane) with hydrogen as the carrier gas. Lipids were extracted by melting the subcutaneous fat in a microwave oven and then filtering and dissolving it in hexane. A total amount of 2783 samples from several campaigns were considered. Using the triacylglycerols as chemical descriptors, a study on the discriminating power to differentiate samples according to the pig feeding type and system was performed. With this aim, pattern recognition techniques, such as linear discriminant analysis (LDA) and multilayer perceptron artificial neural networks (MLPANN), have been used. ANN performed better than LDA, with a mean prediction ability of approximately 97% in the differentiation of fattening diets such as Montanera, extensive Cebo and intensive Cebo. In the case of including the recebo fattening diet, the model presents a mean performance of 82%. The differentiation of fattening systems has also been achieved by means of ANN, with a mean performance of 96%.

Online estimation of multicomponent heat flux using a system identification technique

In this work system identification techniques are used to map the two-dimensional heat flux into the temperatures through a linear model supported by theoretical and numerical results. The basis of this analysis is a discrete version of the Burggraf Method saying a single component heat flux is a linear combination of the temperatures around the time of its occurrence. Taking the same approach, a linear model (i.e. a linear artificial neural network (ANN)) is employed to estimate a multicomponent heat flux as a linear function of the temperatures. A known heat flux is imposed to the direct model, then the history of heat flux-temperature data are fit to the linear mathematical model (i.e. a linear ANN) using system identification techniques. The achieved model estimates the heat flux based on a series of past and future temperatures and the estimated heat flux components are in a good agreement with the exact ones. Finally, the effect of some important factors on the results is investigated. The proposed solution to inverse heat conduction problems does not need thermophysical and geometrical parameters of the system and is robust against noises. It merely needs some series of heat flux-temperature data from solution of a reliable direct numerical model or experiment.

A Local Linear Neurofuzzy Model for the Prediction of Permeability from Well-log Data in Carbonate Reservoirs

The prediction of permeability in heterogeneous carbonates from well-log data represents a difficult and complex problem. The authors present a new approach for the prediction of permeability. It is based on the recently developed neurofuzzy interpretation of locally linear models, which has led to the introduction of the intuitive incremental learning algorithm referred to as the locally linear model tree. The incremental learning algorithm initializes the model with an optimal linear least-squares estimate and automatically increases the number of neurons in each epoch. The model is optimized for the number of neurons to avoid overfitting and to provide maximum generalization by considering the error index of validation sets during training. The effectiveness of the methodology is demonstrated with a case study in a heterogeneous carbonate reservoir (Fahliyan formation in southwest Iran). Wireline data and core data sets from three wells that are located in the center of the field provide the data for the learning task. Core permeability and well-log data from a forth well provide the basis for model validation. The numerical results of the neurofuzzy modeling method are compared with the results of multiple linear regression and artificial neural network. The results are very satisfactory and open the possibilities for future research and applications.

Development of a stochastic weather generator for the sub-polar North Atlantic

The article presents an approach for creating a computationally efficient stochastic weather generator. In this work the method is tested by the stochastic simulation of sea level pressure over the sub-polar North Atlantic. The weather generator includes a hidden Markov model, which propagates regional circulation patterns identified by a self organising map analysis, conditioned on the state of large-scale interannual weather regimes. The remaining residual effects are propagated by a regression model with added noise components. The regression step is performed by one of two methods, a linear model or artificial neural networks and the performance of these two methods is assessed and compared. The resulting simulations express the range of the major regional patterns of atmospheric variability and typical time scales. The long term aims of this work are to provide ensembles of atmospheric data for applied regional studies and to develop tools applicable in down-scaling large-scale ocean and atmospheric simulations.

ARTIFICIAL NEURAL NETWORK FOR PREDICTING NUTRIENTS CONCENTRATION IN RUNOFF FROM BEEF CATTLE FEEDLOT

Nutrient runoff is an environmental concern. There is a need for a robust method of predicting nutrient concentrations in runoff from feedlots to facilitate management practices. To investigate the feasibility of using multiple linear regression (MLR) and artificial neural network (ANN) in predicting nutrients in runoff, simple water quality paramters (e.g., pH and EC) of runoff from a feedlot in North Dakota, USA, were used to train an ANN. Both models accurately predicted potassium concentration based on inputs pH and electrical conductivity (EC). The ANN approach, however, used in this study gave a better prediction than multiple linear regression models. It may be concluded that ANN may be a useful tool for predicting nutrients concentration in runoff using pH and EC, when expensive and time-consuming analytical data are not available, and this information may be used for implementing measures to minimize environmental concerns.

Application of Soft Computing Methods in Predicting Evapotranspiration

Exact prediction of evapotranspiration is necessary for study, design and management of irrigation systems. In this research, the suitability of soft computing approaches namely, fuzzy rule base, fuzzy regression and artificial neural networks for estimation of daily evapotranspiration has been examined and the results are compared to real data measured by lysimeter on the basis of reference crop (grass). Using daily climatic data from Haji Abad station in Hormozgan, west of Iran, including maximum and minimum temperatures, maximum and minimum relative humidities, wind speed and sunny hours, evapotranspiration was predicted by soft computing methods. The predicted evapotranspiration values from fuzzy rule base, fuzzy linear regression and artificial neural networks show root mean square error (RMSE) of 0.75, 0.79 and 0.81 mm/day and coefficient of determination of (R2) of 0.90, 0.87 and 0.85, respectively. Therefore, fuzzy rule base approach was found to be the most appropriate method employed for estimating evapotranspiration.

Application of multi regressive linear model and neural network for wear prediction of grinding mill liners

The liner of an ore grinding mill is a critical component in the grinding process, necessary for both high metal recovery and shell protection. From an economic point of view, it is important to keep mill liners in operation as long as possible, minimising the downtime for maintenance or repair. Therefore, predicting their wear is crucial. This paper tests different methods of predicting wear in the context of remaining height and remaining life of the liners. The key concern is to make decisions on replacement and maintenance without stopping the mill for extra inspection as this leads to financial savings. The paper applies linear multiple regression and artificial neural networks (ANN) techniques to determine the most suitable methodology for predicting wear. The advantages of the ANN model over the traditional approach of multiple regression analysis include its high accuracy.