Performance Of Feed-forward Neural Network Research Articles

Classification between Elderly Voices and Young Voices Using an Efficient Combination of Deep Learning Classifiers and Various Parameters

The objective of this research was to develop deep learning classifiers and various parameters that provide an accurate and objective system for classifying elderly and young voice signals. This work focused on deep learning methods, such as feedforward neural network (FNN) and convolutional neural network (CNN), for the detection of elderly voice signals using mel-frequency cepstral coefficients (MFCCs) and linear prediction cepstrum coefficients (LPCCs), skewness, as well as kurtosis parameters. In total, 126 subjects (63 elderly and 63 young) were obtained from the Saarbruecken voice database. The highest performance of 93.75% appeared when the skewness was added to the MFCC and MFCC delta parameters, although the fusion of the skewness and kurtosis parameters had a positive effect on the overall accuracy of the classification. The results of this study also revealed that the performance of FNN was higher than that of CNN. Most parameters estimated from male data samples demonstrated good performance in terms of gender. Rather than using mixed female and male data, this work recommends the development of separate systems that represent the best performance through each optimized parameter using data from independent male and female samples.

Application of feed forward neural network model to predict the limiting current of tin magneto electrodeposition

Predicting the value of Tin Magneto electrodeposition (MED) is very important since the optimum mass transport occurred at the limiting current. The MED limiting current able to detect using electroanalytical chemistry, but this method is expensive; it needs some method, which able to predict the limiting current of tin MED. However, predicting the limiting current under magnetic field effect is more complicated due to the highly nonlinear characteristic and complicated of its multiple inputs single-output (MISO) system. The nonlinear model that able to predict the limiting current of tin MED is Artificial Neural Networks (ANNs). One of the ANNs which able to simulate the Multiple-Input-Single-Output (MISO) model is the Feed Forward Neural Network (FFNN). In this work, MISO FFNN will model a matrix data set with six variable inputs and one output. The data was obtained from the results of the experiments using electroanalytical chemistry. The output of this model is the limiting current of tin MED, meanwhile, the inputs are by the concentration of tin (Sn2+) in the electrolyte (C), viscosity(v), diffusion coefficient (D), area of the electrode (A), the number of electroactive species (n) and magnetic field strength (B). To get the best model, the performance of FFNN was tested with three variations of the algorithm (Levenberg-Marquardt, Bayesian Regularization, and Scaled Conjugate Gradient) and ten variations of the number of neurons (10, 15, 20, 25, 30, 35, 40, 45 and 50). The best model obtained for this MISO FFNN model is which uses the Levenberg-Marquardt algorithm and the highest number of neurons (50 neurons).

Comparison of simple feedforward neural network, recurrent neural network and ensemble neural networks in phishing detection

The internet has been one of the greatest advancements in technologies. It has brought many advantages to today’s society in many domains such as e-commerce, entertainment and supply chain, amongst others. However, it is also a double-edged sword which has brought many threats to the computer systems and devices known as cyber-attack, and one of these threats would be phishing attack. A phishing attack is where the scammer tries to impose or clone the legitimate email or website in order to deceive the victim to key in their personal information such as username and password. Phishing attack has been one of the most common attacks that happens every day on the Internet especially through email. Many methods have been devised to encounter phishing attack, and one of approaches is through training and monitoring team. These manual approaches, however, are user’s experience-dependent and cost-inefficient. Therefore, many have adopted AI approach instead to detect phishing attack. This paper is one of the many efforts to detect the phishing attack through email by adopting AI method. The objective of this paper is to investigate the performance of feedforward neural network, recurrent neural network and ensemble neural network in phishing email detection. The result of this comparison is empirically evaluated.

Prediction of compressive and flexural strengths of jarosite mixed cement concrete pavements using artificial neural networks

In this paper, an attempt has been made to apply and compare the prediction capability of two variants of Artificial Neural Networks: feed-forward neural network (FFNN) and the radial basis function network (RBFN) for modelling the flexural and compressive strengths of jarosite mixed cement concrete for pavements. The compressive strength and the flexural strength are dependent upon a total of eight inputs. Their values are experimentally determined from the specimens containing 0%, 10%, 15%, 20% and 25% of the jarosite as a partial replacement to cement in M40 concrete mix. These specimens had undergone a curing for 3, 7, 28, 90, 180 and 365 days providing the inputs–outputs experimental data for the 90 observations. The simulation experiments showed that the performance of FFNN is found to be better than that of the RBFN model as the former model provided lesser values of the performance indicators such as Mean Average Error and Mean Squared Error. Further, the FFNN model required a lesser number of parameters to be tuned during the training as compared to the RBFN model.

Parkinson‟s Disease Prediction using Modified Gauss-Newton Method in Feed-Forward Neural Network

Parkinson’s disease (PD) is a brain disorder, characterized by the relapse of the nervous system that spreads gradually in the body. The symptom of PD includes a loss of body control (moderate movement, resting tremors, postural shakiness etc.). So, it is required to detect at an early stage. Machine learning (ML) deals with a variety of probabilistic methods to identify a pattern in a dataset. Therefore, the research is carried out to predict the PD using Multilayer Feed-Forward Neural Network. In Neural Network (NN), weight optimization performed at each layer that plays a major role in the prediction. First-order weight optimization techniques are slow in computation because they reduce the sum of square error using parameter updating in the steepest descent way. In proposed work, a modified recursive Gauss-Newton method is used to optimize the weights for speed up the performance of Feed-Forward NN. This approach is compared with widely used optimization techniques. The Proposed method found better than other techniques and performs fast in Apache Spark than R-Studio framework.

A Biological Mechanism Based Structure Self-Adaptive Algorithm for Feedforward Neural Network and Its Engineering Applications

Feedforward neural network (FNN) is an information processing system that simulates human brain function to a certain extent by referring to the structure of a biological neural network and the working mechanism of biological neurons. As the most important part, the architecture of FNN essentially influences its application performance. This paper proposed a structure self-adaptive algorithm for FNN (SSAFNN) based on biological mechanism. First, self-adaptive neuron growing and pruning indexes are proposed based on the idea of biological neuron grow factor and neuron competition, respectively. The FNN structure is dynamically adjusted according to the growing and pruning indexes of hidden neurons. Second, the connect weights of FNN are automatically adjusted during the self-organizing process and trained by gradient descent method during the learning process. Third, the theoretical analysis is given that this proposed algorithm not only optimized the network structure but also ensured the convergence and performance of FNN. The proposed SSAFNN is tested on the benchmark problems in the field of classification and prediction and is applied in the engineering problems of anchor bolt non-destructive testing and wastewater effluent ammonia nitrogen predicting. The experiment results reveal the good performance and potentiality of SSAFNN in industrial applications.

Applicability of a processes-based model and artificial neural networks to estimate the concentration of major ions in rivers

Modelling is an alternative solution to reduce the cost of water quality monitoring. Commonly, concentration of pollutants is estimated based on limited sampling information. Concentration of ions in rivers can be estimated using modelling strategies that involve statistics and artificial intelligence as well as the understanding of physical processes. Therefore, the performance of feedforward neural networks that employs the Levenberg-Marquardt optimization method was compared to the PPBM recently proposed. Both ANN and PPBM were used to estimate the concentration of major ions (Na+, K+, Mg2+, Ca2+, HCO3−, SO42−, Cl−, and NO3−) in river water based on pH, alkalinity, and temperature. Root-mean-square error and Pearson correlation coefficient (R) together with its p-value were used to evaluate the quality of results of both models. The ANN model provides better estimates compared to the PPBM in most cases. However, the PPBM has the possibility to evaluate its predictions by using the difference between the estimated and measured electrical conductivity. If the predictions are not good the PPBM can be recalibrated, whereas the ANN model is limited in this respect. Another disadvantage of ANN models is that they are developed based on historical data and if limited data are available, such models cannot be used. This latter disadvantage makes the PPBM superior in developing countries, where often little or no consistent historical data exist.

HALF-WAVE SEGMENT FEATURE EXTRACTION OF EEG SIGNALS OF PATIENTS WITH DEPRESSION AND PERFORMANCE EVALUATION OF NEURAL NETWORK CLASSIFIERS

A detailed understanding of key signal characteristics has enabled the use of artificial neural networks (ANN) for feature detection and classification of EEG signals in clinical research. The present study is performed to classify EEG signals of normal and depression patients with wavelet parameters as key input features. The characteristics of depression cannot be made out by visual inspection of EEG records unlike epilepsy which is well characterized by sudden recurrent and transient waveforms. In this study, a comparison is made between the performance of feedforward neural network (FFNN) and probabilistic neural network (PNN) while classifying the EEG signals of normal and depression patients. Classification capabilities of both the methods are validated with the EEG recordings from 30 normal controls and 30 depression patients. One-way ANOVA provided a statistical significant difference between the two classes of EEG signals recorded. Preprocessing for feature extraction is done using discrete wavelet transform (DWT). The time domain and relative wavelet energy (RWE) features calculated from the sub-bands are given as a set of input to the neural network. Another set of feature used independently for training the network is the wavelet entropy (WE). The FFNN achieved a classification accuracy of 100% and PNN gave an accuracy of 58.75% with time domain and wavelet energy as the input features. With wavelet entropy as the input feature, FFNN further showed 98.75% classification accuracy while PNN gave an accuracy of only 46.5%. The results indicate that FFNN with the given input features is more suitable for the classification of EEG signals with mood changing depressive disorders.

Use of Diverse Chemometric and Validation Methods to Accurately Predict Human Urotensin-II Receptor Antagonist Activity.

Despite being identified as the most potent receptor related to vasoconstriction, human urotensin-II receptor (hUT) has not been fully explored as a target for the treatment of cardiovascular diseases. In view of this and with an aim to identify precise structural requirements for binding of hUT antagonists, we endeavoured to develop, for the first time, multivariate QSAR models using chemometric methods like partial least squares (PLS) and feed-forward neural network (FFNN). A set of 48 pyrrolidine derivatives having hUT binding affinity was used for multivariate model development. The accuracy and predictability of the developed models was evaluated using crossvalidation. The PLS model showed good correlation between selected descriptors and Ki values (r(2) =0.745 and r(2) (CV) =0.773). However, the predictive performance of FFNN was better than the PLS technique with r(2) =0.810. The study clearly suggests the role of lipophilic and steric descriptors in the ligand-hUT interactions. The QSAR models generated can be successfully extended to predict the binding affinities and for the effective design of novel hUT antagonists.

Hybrid evolutionary techniques in feed forward neural network with distributed error for classification of handwritten Hindi ‘SWARS’

In this work, the performance of feedforward neural network with a descent gradient of distributed error and the genetic algorithm (GA) is evaluated for the recognition of handwritten ‘SWARS’ of Hindi curve script. The performance index for the feedforward multilayer neural networks is considered here with distributed instantaneous unknown error i.e. different error for different layers. The objective of the GA is to make the search process more efficient to determine the optimal weight vectors from the population. The GA is applied with the distributed error. The fitness function of the GA is considered as the mean of square distributed error that is different for each layer. Hence the convergence is obtained only when the minimum of different errors is determined. It has been analysed that the proposed method of a descent gradient of distributed error with the GA known as hybrid distributed evolutionary technique for the multilayer feed forward neural performs better in terms of accuracy, epochs and the number of optimal solutions for the given training and test pattern sets of the pattern recognition problem.

Analysis of speedup as function of block size and cluster size for parallel feed-forward neural networks on a Beowulf cluster.

The performance of feed-forward neural networks trained with the backpropagation algorithm on a dedicated Beowulf cluster is analyzed. The concept of training set parallelism is applied. A new model for run time and speedup prediction is developed. With the model the speedup and efficiency of one iteration of the neural networks can be estimated as a function of block size and cluster size. The model is applied to three example problems representing different applications and network architectures. The estimation of the model has a higher accuracy than traditional methods for run time estimation and can be efficiently calculated. Experiments show that speedup of one iteration does not necessarily translate to a shorter training time toward a given error level. To overcome this problem a heuristic extension to training set parallelism called weight averaging is developed. The results show that training in parallel should only be done on clusters with high performance network connections or a multiprocessor machine. A rule of thumb is given for how much network performance of the cluster is needed to achieve speedup of the training time for a neural network.

Learning improvement of neural networks used in structural optimization

The performance of feed-forward neural networks can be substantially impaired by the ill-conditioning of the corresponding Jacobian matrix. Ill-conditioning appearing in feed-forward learning process is related to the properties of the activation function used. It will be shown that the performance of the network training can be improved using an adaptive activation function with a properly updated gain parameter during the learning process. The efficiency of the proposed adaptive procedure is examined in structural optimization problems where a trained neural network is used to replace the structural analysis phase and capture the necessary data for the optimizer. The optimizer used in this study is an algorithm based on evolution strategies.

Improving the Performance of Feedforward Neural Networks by Noise Injection into Hidden Neurons

The generalization ability of feedforward neural networks (NNs) depends on the size of training set and the feature of the training patterns. Theoretically the best classification property is obtained if all possible patterns are used to train the network, which is practically impossible. In this paper a new noise injection technique is proposed, that is noise injection into the hidden neurons at the summation level. Assuming that the test patterns are drawn from the same population used to generate the training set, we show that noise injection into hidden neurons is equivalent to training with noisy input patterns (i.e., larger training set). The simulation results indicate that the networks trained with the proposed technique and the networks trained with noisy input patterns have almost the same generalization and fault tolerance abilities. The learning time required by the proposed method is considerably less than that required by the training with noisy input patterns, and it is almost the same as that required by the standard backpropagation using normal input patterns.

Curvature-driven smoothing: a learning algorithm for feedforward networks

The performance of feedforward neural networks in real applications can often be improved significantly if use is made of a priori information. For interpolation problems this prior knowledge frequently includes smoothness requirements on the network mapping, and can be imposed by the addition to the error function of suitable regularization terms. The new error function, however, now depends on the derivatives of the network mapping, and so the standard backpropagation algorithm cannot be applied. In this letter, we derive a computationally efficient learning algorithm, for a feedforward network of arbitrary topology, which can be used to minimize such error functions. Networks having a single hidden layer, for which the learning algorithm simplifies, are treated as a special case.