Inputs Of Artificial Neural Network Research Articles

Artificial Neural Network for Prediction of Venous Thrombosis Recurrence

▪Background: Prediction of venous thrombosis recurrence is still a challenge. Vienna and Dash scores are widely used for this task, however its results represent only a percentage risk and does not considerate several factors that may influence recurrent thrombosis. In this context, artificial neural networks (ANN) could be a powerful tool to relate the clinical and laboratorial parameters with the risk of venous thrombosis recurrence.Aims: To obtain a ANN model which calculate whenever a patient may have rethrombosis based in his clinical and laboratorial parameters.Methods: Data of 240 patients with a history of Deep Vein Thrombosis (DVT) were included until one year after the thrombotic episode. Forty four clinical and laboratorial parameters were considered: gender, age, site of thrombosis, pulmonary embolism, proximal or distal lower limb thrombosis, provoked or unprovoked thrombosis, D-dimer, FV Leiden, G20210A prothrombin gene mutation, factor VIII, protein C, protein S, antithrombin, lupus anticoagulant, anticardiolipin, anti β2-GPI, antiphospholipid syndrome, body mass index, tabagism, hormonal therapy (for women), hemogram (leukocyte, platelet, hemoglobin, erythrocyte, hematocrit, RDW, MPV), HDL and LDL cholesterol, triglycerides, glycemia, creatinine, C-reative protein, diabetes mellitus, arterial hypertension, dyslipidemia, liver and/or renal failure, cancer, residual venous thrombus on US doppler, sickle-cell disease, anticoagulant treatment including drug and duration. All these variables were used as ANN inputs. The occurrence (or not) of rethrombosis was used as output in a yes/no (1/-1) form. The data were randomly divided into train (60 %), test (15 %), and validation (15 %) data. ANNs with one and two hidden layers were tested. The number of neurons varied from 5 to 35 at the first hidden layer and from 5 to 20 at the second hidden layer. ANNs were trained using the Powell method, minimizing the sum of quadratic errors between the calculated and the experimental response.Results: The ANN that gave the best result contained 44 neurons in the input layer, 5 neurons in the intermediate layer and 1 neuron in the output layer. Figure 1 shows the correlation between the calculated and the experimental outputs, which are very high. The overall model correlation was 0,97. Figure 2 confirms this result: the absolute error between the experimental and the calculated value is very low, and are very near to zero in most cases. The mean average deviation obtained was 2.5 %.Conclusions: Our results suggest that using a set of clinical and laboratorial data to construct an ANN model could be an efficient tool to predict the recurrence of venous thrombosis. This finding must be repeated in the same group of study including a higher number of patients, and be validate in other populations to be used in clinical practice. We also intend to select the variables that could be related to recurrence of venous episode. [Display omitted] [Display omitted] DisclosuresNo relevant conflicts of interest to declare.

Prediction of HIV-1 protease cleavage site using a combination of sequence, structural, and physicochemical features.

BackgroundThe human immunodeficiency virus type 1 (HIV-1) aspartic protease is an important enzyme owing to its imperative part in viral development and a causative agent of deadliest disease known as acquired immune deficiency syndrome (AIDS). Development of HIV-1 protease inhibitors can help understand the specificity of substrates which can restrain the replication of HIV-1, thus antagonize AIDS. However, experimental methods in identification of HIV-1 protease cleavage sites are generally time-consuming and labor-intensive. Therefore, using computational methods to predict cleavage sites has become highly desirable.ResultsIn this study, we propose a prediction method in which sequence, structural, and physicochemical features are incorporated in various machine learning algorithms. Then, a bidirectional stepwise selection algorithm is incorporated in feature selection to identify discriminative features. Further, only the selected features are calculated by various encoding schemes and used as input for decision trees, logistic regression, and artificial neural networks. Moreover, a more rigorous three-way data split procedure is applied to evaluate the objective performance of cleavage site prediction. Four benchmark datasets collected from previous studies are used to evaluate the predictive performance.ConclusionsExperiment results showed that combinations of sequence, structure, and physicochemical features performed better than single feature type for identification of HIV-1 protease cleavage sites. In addition, incorporation of stepwise feature selection is effective to identify interpretable biological features to depict specificity of the substrates. Moreover, artificial neural networks perform significantly better than the other two classifiers. Finally, the proposed method achieved 80.0% ~ 97.4% in accuracy and 0.815 ~ 0.995 evaluated by independent test sets in a three-way data split procedure.Electronic supplementary materialThe online version of this article (doi:10.1186/s12859-016-1337-6) contains supplementary material, which is available to authorized users.

Metode Jaringan Syaraf Tiruan Untuk Prediksi Performa Mahasiswa Pada Pembelajaran Berbasis Problem Based Learning (PBL)

In order to improve academic quality in higher education, students’ performance evaluation is becoming important. To prevent increasing failure rate in the course, we need a system that is capable of predicting student’s performance in the end of the course. The research used several factors that are considered to affect students' performance on Problem Based Learning (PBL), such as students’ demography, students’ prior knowledge and group heterogeneity. The method used in the study was Artificial Neural Network (ANN) with backpropagation training algorithm. Total 8 neurons were used as inputs for ANN which were obtained from gender variable (2 neurons), age variable (1 neuron), students’ average knowledge variable (1 neuron), students’ average skill variable (1 neuron) and group heterogeneity variable (3 neurons). Several different ANN architecture were tested in the study using 2, 7 and 12 hidden neurons respectively. Each architecture was trained using various different training parameters in order to find the best ANN architecture. Dataset used in the research were obtained from Academic Information System in Faculty of Dentistry Unissula which contained Adult and Elderly Diseases Course’s participants from year 2009 to 2013. The ANN output were numeric values which represented students’ performance in Adult and Elderly Diseases Course. The output of this study is a system that is able to predict the student performance in block course. The result shows that using 7 hidden neurons in the network combining with 0.5 ,0.1 and 9000 for learning rate, momentum and epoch respectively, were the best ANN architechture and parameters in the study. The MSE obtained from validation test was 0,011926 with correlation coefficient (R) 0,796879. The prediction system are expected to help faculty and academic evaluation team to conduct actions to improve student’s academic performance and prevent them from failure in the course.

Predicting Near‐Surface Moisture Content of Saline Soils from Near‐Infrared Reflectance Spectra with a Modified Gaussian Model

Core Ideas Moisture estimates from near‐infrared reflectance are improved with a modified Gaussian model. Artificial neural network simulations further improve moisture prediction accuracy. The proposed method yields better results than standard reflectance moisture indices. Near‐surface soil moisture is a spatially highly heterogeneous state variable that affects the hydrologic partitioning of precipitation, net ecosystem exchange, and land–atmosphere interactions. In concert with the rapid advancement of large‐scale hyperspectral remote sensing technology over the last decade, numerous methods relating reflectance to near‐surface soil moisture have been proposed. Although there is evidence that surface reflectance is conjointly controlled by moisture and salt content, little is known about the impact of soil salinity. To study the effects of soil salinity, near infrared to short‐wave infrared reflectance spectra were measured in a controlled laboratory environment for samples representing a wide range of salinity levels. A previously proposed surface moisture prediction method based on geometrical attributes of an inverted Gaussian (IG) function fitted to hyperspectral reflectance curves was revisited. Improvements were proposed by first modifying the geometrical attributes of the IG applied to predict near‐surface soil moisture and by concurrently considering multiple geometrical attributes of the IG function as input for trained artificial neural networks (ANNs). Although previously applied linear regression models that relate a single geometrical parameter to soil moisture failed to satisfactorily predict independently measured surface soil moisture, considerable improvements in prediction accuracy were achieved with both geometrical attribute modification and ANN simulations.

ANN Prediction of Metabolic Syndrome: a Complex Puzzle that will be Completed.

The diagnosis of metabolic syndrome (MetS) has a leading role in the early prevention of chronic disease, such as cardiovascular disease, type 2 diabetes, cancers and chronic kidney disease. It would be very greatful that MetS diagnosis can be predicted in everyday clinical practice. This paper presents artificial neural network (ANN) prediction of the diagnosis of MetS that includes solely non-invasive, low-cost and easily-obtained diagnostic methods. This solution can extract the risky persons and suggests complete tests only on them by saving money and time. ANN input vectors are very simple and contain solely non-invasive, low-cost and easily-obtained parameters: gender, age, body mass index, waist-to-height ratio, systolic and diastolic blood pressures. ANN output is M e t S-coefficient in true/false form, obtained from MetS definition of International Diabetes Federation (IDF). ANN training, validation and testing are conducted on the large dataset that includes 2928 persons. Feed-forward ANNs with 1-100 hidden neurons were considered and an optimal architecture were determinated. Comparison with other authors leads to the conclusion that our solution achieves the highest positive predictive value P P V = 0.8579. Further, obtained negative predictive value N P V = 0.8319 is also high and close to PPV, which means that our ANN solution is suitable both for positive and negative MetS prediction.

Prediction of storage efficiency on CO2 sequestration in deep saline aquifers using artificial neural network

This study presents the application of artificial neural network (ANN) to predict storage efficiency of CO2 sequestration in deep saline aquifers. To a create training database used as input and output neurons in ANN, sensitivity analysis of parameters considering the aquifer characteristics was performed by numerical simulation. Based on the analysis, the factors and their ranges influencing CO2 sequestration in deep saline aquifers were determined and 150 representative realizations used as a training database were generated with trapping indices of the residual CO2 and solubility trapping mechanisms. The ANN model was designed with optimum architecture minimizing the mean squared error (MSE) for testing data set and it was tested with validation samples. The results showed that the proposed ANN model had a high prediction performance with a high coefficient of determination (R2) of over 0.99 on comparing with the target values, a low MAPE (mean absolute percentage error) of 1.26%, and RMSE (root mean square error) of 0.41 for TEI (total trapping efficiency index). As a field application, the model has also been evaluated with the field scale data on the Gorae V structure in Block VI-1, Korea. The results of prediction were well matched with the targeted values and accuracy between the ANN predictions and field scale data was achieved with a high coefficient of determination (R2) of more than 0.96, a low MAPE of 4.40%, and RMSE of 3.58 for TEI. From these results, it is believed that the newly developed ANN model can predict storage efficiency with high accuracy and it can also be considered as a useful and robust tool to evaluate the feasibility of CO2 sequestration in deep saline aquifers.

Artificial neural-network-based fault location for power distribution lines using the frequency spectra of fault data

This study presents an artificial neural-network (ANN)-based digital fault classification and location algorithm for medium voltage (MV) overhead power distribution lines with load taps and embedded remote-end source. The algorithm utilizes frequency spectra of voltage and current samples which are recorded by the digital relay at the substation. In the algorithm, to extract useful information for ANN inputs, the frequency spectral analysis of voltage and current waveforms has been carried out using Fast Fourier Transform. To classify and locate the shunt faults on an MV distribution system, a multilayer perceptron neural network (MLPNN) with the standard back-propagation technique has been used. A 34.5 kV overhead distribution system has been simulated using MATLAB/Simulink, and the results are used to train and test the ANNs. The technique takes into account all the practical aspects of real distribution system, such as errors, originated from instrument transformers and interface. The ANN-based fault location technique has been extensively tested for a realistic model and gives satisfactory results for radial overhead distribution systems with load taps and in the presence of remote-end source connection.

Applicability of Artificial Neural Network and Nonlinear Regression to Predict Mechanical Properties of Equal Channel Angular Rolled Al5083 Sheets

Equal channel angular rolling (ECAR) is a severe plastic deformation (SPD) process in order to achieve ultrafine-grained (UFG) structure. In this paper, the mechanical properties of ECAR process using artificial neural network (ANN) and nonlinear regression have been illustrated. For this purpose, a multilayer perceptron (MLP) based feed-forward ANN has been used to predict the mechanical properties of ECARed Al5083 sheets. Channel oblique angle, number of passes and the route of feeding are considered as ANN inputs and tensile strength, elongation and hardness are considered as the outputs of ANN. In addition, the relationship between input parameters and mechanical properties were extracted separately using nonlinear regression method. Comparing the outputs of models and experimental results shows that models used in this study can predict and estimate mechanical properties appropriately. Where, the performance of ANN model is better than the correlations to predict mechanical properties. Finally, the developed outputs of neural network model are used to analyze the effects of input parameters on tensile strength, elongation and hardness of ECARed Al5083 sheets.

Artificial intelligence approach based on near-infrared spectral data for monitoring of solid-state fermentation

This work aimed to establish a chemometric technique for quantifying amylase and protease activities as well as protein concentration in aqueous extracts of Rhizopus microsporus var. oligosporus obtained via solid-state fermentation (SSF). The kinetics of four agro-industrial wastes (wheat bran, soybean meal, type II wheat flour and sugarcane bagasse) were studied for 144h, along with two different sets of their ternary mixtures, at a constant fermentation time of 120h, to obtain primary data (biochemical parameters as well as near-infrared (NIR) spectral data). Then, models such as artificial neural network (ANN) and partial least squares (PLS) were calibrated to predict biochemical parameters using the spectral data. Primary data and three methods of preprocessing data – first, second and third derivatives – were assessed as inputs for both chemometric tools. The third derivative, that is, spectral pre-processing plus an optimized ANN, showed the least relative errors (<8.3%±10.5%). The third-derivative spectrum was found to be suitable as the ANN input data for monitoring amylase and protease activities and protein concentration in the SSF under study. The proposed methodology can serve as a foundation for at-line sensor development and decrease the time and cost of bioprocess development using Rhizopus microsporus var. oligosporus.

Computational fluid dynamics assists the artificial neural network and genetic algorithm approaches for thermal and flow modeling of air-forced convection on interrupted plate fins

ABSTRACTThis study reports the application of Computational Fluid Dynamics (CFD) as a data provider for Artificial Neural Networks (ANNs). Four interrupted plate fins with different geometric parameters were studied experimentally. The CFD modeling was undertaken and the simulation results were verified using the experimental data. After validating the model, more fins with various geometries were modeled. The numerically validated data were used for developing two ANNs. Reynolds number and geometric parameters were determined as ANN inputs, and Nusselt number (Nu) and friction factor (f) were outputs. Moreover, the ANNs were compared to genetic algorithm-based correlations and the ANNs appeared more accurate than the correlations.

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.

Investigation of using 60Co source and one detector for determining the flow regime and void fraction in gas–liquid two-phase flows

In this study, a simple detection system comprised of one 60Co source and just one NaI detector was investigated in order to identify flow regime and measure void fraction in gas–liquid two phase flows. For this purpose, 3 main flow regimes of two-phase flows including stratified, homogenous and annular with void fractions in the range of 5–95% were simulated by Monte-Carlo N Particle (MCNP) code. At first step, 3 features (count under full energy peaks of 1.173 and 1.333MeV, and count under Compton continuum) were extracted from registered gamma spectrum. These 3 extracted features were used as inputs of artificial neural network (ANNs). A primary network was trained for identifying the flow regimes, but after testing many different structures, it was found that just two regimes of stratified and annular could be completely identified from each other. After identifying the mentioned two flow regimes by the first ANN, two specific ANNs were also implemented for predicting the void fraction. Using the proposed method in this work, void fraction percentages were predicted with a mean relative error (MRE) of less than only 0.42%. Using fewer detectors is of advantage in industrial nuclear gauges, because of reducing economical expenses and also simplicity of working with these systems.

Benefits of metamodel-reduction for nonlinear dynamic response analysis of damaged composite structures

In this paper, a novel method for damage prediction and dynamic behaviour analysis of laminate composite structures is proposed and investigated. The dynamic behaviour of transversely isotropic layers is expressed through elasticity coupled with damage using a phenomenological macro-model for cracked composite structures made of polymer reinforced with long glass fibres. The damage is fully described by a single scalar variable whose evolution law is expressed through the maximum dissipation principle. Using the classical linear Kirchhoff–Love theory of plates and considering the damage-induced nonlinearity, the obtained nonlinear dynamic equations are solved in time domain using a Newmark algorithm. To reduce the computational costs, a metamodel-reduction for damage localization and quantification is proposed where the Artificial Neural Networks (ANNs) and Craig–Bampton reduction methods are combined. Extracted stresses from finite element analysis are used as input for a feed-forward ANNs to estimate the damage severity. The Craig–Bampton reduction method is introduced as a Component Mode Synthesis (CMS) to investigate the case of assembled structures locally damaged. Numerical simulations show that the damage modifies significantly the dynamic properties restricted to the eigenfrequencies reduction. The designed feed-forward ANNs was verified and it provides promising results regarding severity and location of the damage. Moreover, the trained ANNs provide a quick response for damage level prediction in online procedure which permits to significantly reduce the computational costs.

Optimization of Melatonin Dissolution from Extended Release Matrices Using Artificial Neural Networking.

Efficacy of melatonin in treating sleep disorders has been demonstrated in numerous studies. Being with short half-life, melatonin needs to be formulated in extended-release tablets to prevent the fast drop of its plasma concentration. However, an attempt to mimic melatonin natural plasma levels during night time is challenging. In this work, Artificial Neural Networks (ANNs) were used to optimize melatonin release from hydrophilic polymer matrices. Twenty-seven different tablet formulations with different amounts of hydroxypropyl methylcellulose, xanthan gum and Carbopol®974P NF were prepared and subjected to drug release studies. Using dissolution test data as inputs for ANN designed by Visual Basic programming language, the ideal number of neurons in the hidden layer was determined trial and error methodology to guarantee the best performance of constructed ANN. Results showed that the ANN with nine neurons in the hidden layer had the best results. ANN was examined to check its predictability and then used to determine the best formula that can mimic the release of melatonin from a marketed brand using similarity fit factor. This work shows the possibility of using ANN to optimize the composition of prolonged-release melatonin tablets having dissolution profile desired.

Fault Identification in the Stator Winding of Induction Motors Using PCA with Artificial Neural Networks

Three-phase induction motors are the main element of electrical into mechanical energy conversion applied in the industries. Due to its constant usage, added to adversities such as thermal, electrical and mechanical, these motors can be damaged causing unexpected process losses. Among the drawbacks of occurrences commonly presented for this equipment, approximately 37 % are related to short circuit in the stator coils. Hence, this article proposes an alternative approach for stator fault identification in induction motors through the discretization of the current signal, in the time domain, applying a variable optimization technique of principal components analysis (PCA) and artificial neural networks (ANNs) types multilayer perceptron (MLP) and radial basis function. Experimental results are presented with data gathered from an experimental workbench, considering various supply conditions and also under a wide load variation, by using the amplitude of the current signals in the time domain. Moreover, the MLP network presented the best results and the PCA technique provided a considerable reduction in the number of ANNs inputs, and in general, the classification results were comparable to the results obtained when the networks inputs considered the technique employing downsampling of 30 points to represent the current signals using half-cycle of the waveform.

Day-ahead forecasting of solar photovoltaic output power using multilayer perceptron

Penetration of grid-connected photovoltaic systems can be increased substantially by devising area-specific power output forecasting methods. Meteorological conditions of the area are decisive for solar plant management and electricity generation. This paper estimates and forecasts the profile of power output of a grid-connected 20-kWp solar power plant in a reputed manufacturing industry located in Tiruchirappalli, India, using artificial neural networks (ANNs). A multilayer perceptron-based ANN model is proposed for day-ahead forecasting of the power generation. An experimental database comprising of each day's solar power output and atmospheric temperature for a period of 70 days has been used for training the ANN. Various training algorithms, transfer functions, and learning rules in the hidden layers/output layers were employed on the database of 11,200 patterns in order to obtain the best mapping between the ANN's inputs and outputs. Statistical error analysis in terms of mean absolute percentage error calculated on the 24-h-ahead forecasting results is presented. Analysis of the variations in network forecasting performance caused by changing the neuron functional parameters has been carried out. The results are also utilized for load scheduling operations of the industrial grid for the next day. Reliable area-specific solar power production map can help in power system scheduling and investment productivity.

Utilização de Redes Neurais Artificiais em Alertas Hidrológicos: Estudo de Caso na Bacia do Rio Claro em Caraguatatuba, Estado de São Paulo

The river flow prediction of a hydrological basin with natural disaster risk, such as floods and flash floods, is an important feature of early warning programs. This work presents an approach based on the Artificial Neural Networks (ANNs) to predict (neuroforecast) the flow of the Claro River in Caraguatatuba-SP. The observed data of this hydrological basin were used to perform the training, test and validation of the neural networks. The ANN inputs are constituted by n past observed precipitation data and n-1 observed flow data. However, the output of the ANN is composed by n-ith calculated flow data. The choice of the input number (the quantity of past observed data) was made taking into account the following metrics: the NASH coefficient, which is calculated on the temporal data of the network response; and a set of indexes related to the providing an early warning when the estimated flow exceeds a critical flow. Based on performance metrics, the chosen ANN has a good adjustment to the observed flow data (NASH = 0.77) and good ability for providing an early warnings (efficiency of 0.91).

Artificial Neural Networks and Thermal Image for Temperature Prediction in Apples

The inability to correctly implement and safeguard a product cold chain leads to premature product spoilage and increased product waste. Special care is required to both implement and monitor the cold chain for perishable goods in order to preserve them. Many technologies are available on the market today with varying levels of success. This article presents a new technique, namely thermal imaging predicts surface temperature over a pallet of apples whilst comparing packaging (plastic boxes and cardboard boxes). This temperature data was then introduced as an input in artificial neural network (ANN) software to estimate the temperature across the entire pallet. Results obtained (root mean squared error [RMSE]) indicate that the estimation with plastic boxes has an error of 0.41 °C whilst the error, taking as a reference the surface temperature, would be RMSE 2.14 °C. In the case of cardboard boxes, the estimation error is 0.086 °C whilst only taking into account the thermal image, data would be RMSE 3.56 °C. This article proves the concept of the possibility of temperature monitoring by ANN through thermal imaging technology.

Reliability forecasting models for electrical distribution systems considering component failures and planned outages

Many utilities in developing countries are investing in installation and renewing of Electrical Distribution System (EDS) components, such as overhead lines, cables and switching devices, to improve the EDS reliability and meet the rapid increase of load demand. In the beginning stage of investment, it is very difficult to evaluate the EDS reliability by using traditional methods due to EDS topology not being fully determined. This paper presents a comprehensive model for forecasting EDS reliability, which is built separately into two parts, i.e. the models for EDS failures and planned outages. Firstly, a three-layer Artificial Neural Network (ANN) model is proposed to forecast the EDS reliability considering EDS failures. Each neuron in the ANN input layer represents a key influencing factor of EDS failures, which are recognized by Gray Relational Analysis (GRA) method. The proposed ANN is trained using historical reliability data of an EDS. In addition, a planned outage reliability model is also built according to the magnitude of investment and type of planned outage. The priorities of improvement measures can also be obtained using the GRA to improve the EDS reliability. Case studies of practical EDSs illustrate the efficiency and applicability of the proposed techniques.

Performance evaluation of neural network, support vector machine and random forest for prediction of donor splice sites in rice

Prediction of splice sites plays an important rolein predicting the gene structure. Rice being one of the major cereal crops, continuous improvement is possible with the prediction of unknown genes associated with complex traits. Machine learning techniques i.e., Artificial Neural Network (ANN) and Support Vector Machine (SVM) have been successfully used for the prediction of splice sites but comparison of their performance has not been made yet to our limited knowledge. Further, Random Forest (RF), another machine learning method, has been successfully used and reported to outperform ANN and SVM in areas other than splice site prediction. In this study we have developed an approach to encode the splice site sequence data of rice into numeric form that are subsequently used as input in ANN, SVM and RF for prediction of donor splice sites. The performances were then evaluated and compared using receiving operating characteristics (ROC) curve and estimate of area under ROC curve (AUC), averaged over 5-fold cross validation. The result reveals that AUC of RF is higher than ANN and SVM which implies that it can be preferred over SVM and ANN in the prediction splice sites.