This paper consists of two models, based on the famous ArtificialNeuralNetwork (ANN) Models: Multi LayeredPerceptron (MLP) model and Cascade Correlation network model developed and compared in their ability to judge the accuracy in results obtained by the diagnosis of the disease in patients. The significance of disease diagnosis by ArtificialNeural techniques is not at all obscure these days. The increasing demand of ArtificialNeural Networks in the field of medicine has proved to show a significantly better performance in medical decision making. These networks are used to diagnose a wide variety of diseases based on the inputs to the model. The diagnoses are made on specific models with information taken from a large number of patients as compared to a single one. These models do not depend on the assumption made by correlation of different variables. One of the proposed techniques involves the training of the Multi LayeredPerceptron (MLP)to recognize a pattern for the diagnosis and prediction of the diseases. For this purpose, we have used various featured inputs on the basis of patients unique like age, sex, marital status, signs and symptoms. On the whole, the Multi LayeredPerceptron (MLP) model has proved to diagnose the diseases of multiple patient outcomes more accurately than Cascade Correlation network model for the validated data. Moreover the results have also significantly demonstrated the suitability of the neuralnetwork models for specifying the disease the patient possesses.
Abstract—One of the fundamental issues in texture classification is the suitable selection combination of input parameters for the classifier. Most researchers used trial and observation approach in selecting the suitable combination of input parameters. Thus it leads to tedious and time consuming experimentation. This paper presents an automated method for the selection of a suitable combination of input parameters for gray level texture image classification. The Artificial Bee Colony (ABC) algorithm is used to automatically select a suitable combination of angle and distance value setting in the Gray Level Co-occurrence (GLCM) matrix feature extraction method. With this setting, 13 Haralick texture features were fed into Multi-layer PerceptronNeuralNetwork classifier. To test the performance of the proposed method, a University of Maryland, College Park texture image database (UMD Database) is employed. The texture classification results show that the proposed method could provide an automated approach for finding the best input parameters combination setting for GLCM which leads to the best classification accuracy performance of binary texture image classification.
MLPs have evolved over the years as a very powerful technique for solving a wide variety of problems. Much progress has been made in improving performance and in understanding how these neural networks operate. In this paper we develop a network and train it for a function and then analyzing the results. Simultaneously, we would like to investigate the effects of changing the number of hidden layer processing elements. In this paper, we focus on the most common neuralnetwork architecture such as multilayer Perceptron. A multilayer Perceptron is a feedforward artificialneuralnetwork model that maps input data samples onto the appropriate number of outputs. An MLP employ a supervised learning technique named as backpropagation for training the network . The backpropagation algorithm is used in layered feed-forward ANNs. This means that the artificial neurons are organized in layers, and send their signals “forward”, and then the errors are propagated backwards. The network receives inputs by neurons in the input layer, and the output of the network is given by the neurons on an output layer. There may be one or more intermediate hidden layers.
An ANN is configured for a particular application, such as pattern recognition or data classification, through a learning process. The true power and advantage of neural networks lies in their ability to represent both linear and non-linear relationships and in their ability to learn these relationships directly from the data being modeled. There are three basic elements of a neuron model. Figure 1 shows the basic elements of an ANN (the perceptron model):
Abstract- Breast cancer is a serious and life threatening disease due to its invasive and infiltrative character and is very commonly found in woman .An abnormal growth of cells in breast is the main cause of breast cancer actually this abnormal growth of cells can be of two types benign (Non- Cancerous) and malignant (Cancerous) ,these types must be diagnosed clearly for proper meditation and for proper treatment. A physician with full of experience and knowledge can deal complex problem in the breast cancer diagnosis process to identify disease but modern medical diagnosis system is totally based on data obtained through clinical and/or other test ,most of the decision related to a patient to find out disease is taken based on these data Better classification of a disease is a very crucial and challenging job , a small error can cause the problem because it is directly related to the life of a human being. In this research work ,various intelligent techniques including supervised ArtificialNeuralNetwork (ANN) ,unsupervised ArtificialNeuralNetwork ,Statistical and decision tree based have been applied to classify data related to breast cancer health care obtained from UCI repository site. The various individual models developed are tested and combined together to form ensemble model .Experimental works were done using MATLAB and SPSS Clementine software obtained results shows that ensemble model is better than individual models accuracy obtained in case of ensemble model is ,which is higher than all individual models ,however counter propogation network (CPN) is a competitive model among all other individual models and accuracy of this model is very near to that is obtained in case of I ensemble model. In order to reduce dimensionality of breast cancer data set a ranking based feature selection technique is applied with best ensemble model ,experimental result show that model has less accuracy with less number of features .Models are also analyzed in terms of other error measures like sensitivity and specificity.
 used ANN for classification of heart murmurs and got classification accuracy of 85%.  shows for wavelet based feature set for a heart murmur and the best result 86.4% came for the MLP ANN classifier. So for classification the BP MLP ANN is preffered as we are also using wavelet based feature set. In the project we used score generated out- put from the ANN to determine the class of the testing sample. 20 Sec long heart sound testing samplea are processed. And from every sample 15 heart cycles are extracted. This generate 15 feature vectors.After normalization these are sent to the trained ANN. It was seen by trial and verify the method that neuralnetwork with 25 hidden nodes is giving the best result. The specifications of NeuralNetwork used in Matlab is given in Figure 3.7.
The use of Artificial Intelligence techniques such as ArtificialNeuralNetwork (ANN) in Travel demand modelling began in 1960. However it wasn’t used for about next three decades in such study due to its weakness, namely the slow response to the modification of inputs despite its extraordinary success at learning or recognising pattern. Neural networks have been used in the transportation demand forecasting by (Chin et al., 1992) and forecasting intercity flows by (Nijkamp et al., 1996). (Rao et al., 1998) has showed ArtificialNeuralNetwork advantages in use for traffic behavioural analysis. Forecasting by ANN is done by minimising an error term indicated as the deviation between input and output through the use of specific training algorithm and random learning rate (Black, 1995; Zhang et al., 1998). The theorem proved by (Hornik et al., 1989) and (Cybenko, 1989) states that a multilayered feed forward neuralnetwork with one hidden layer can approximate any continuous function up to a desired degree of accuracy provided it contains a sufficient number of nodes in the hidden layer thus they can be considered as universal approximates. (Xie et al., 2003) considered two data mining methods, namely learning tree algorithm and neural networks (back propagation) to improve performance of mode choice forecast. (Golias and Karlaftis, 2001) proposed a recursive partitioning methodology for individual mode choice prediction. The methodology is based on tree-structured nonparametric classification technique. (Edara, 2003) developed mode choice models using artificialneural networks. Data mining procedures like clustering are used to process the extracted data.
This paper highlights the fact that how an android app will make it possible for every common man to ease the tedious process of scanning and converting the images into soft copy just on a click even while working offline. It focuses on comparative study and implementation of OCR concepts and principles using ArtificialNeuralNetwork and Nearest Neighbor concepts. It also clearly explains every processing stage in detail and efforts taken to overcome the drawback of existing mobile device OCR applications and limitations of mobile device capabilities so that it generates powerful app which processes things in reduced time and increased accuracy that recognizes almost every different font.
En esta investigación, se aplicó un modelo de red neuronal artificial (ANN), para estimar las condiciones térmicas de las regiones montañosas de Gerania (MG) y de Nafpaktia (MN) en Grecia. La temperatura del aire y la humedad relativa fueron registradas de junio hasta agosto de 2007, en dos sitios seleccionados de cada región estudiada. Datos de los parámetros antes mencionados se usaron para calcular el índice termohi- grométrico (THI), evaluando las condiciones de confort térmico como categorías. El modelo ANN, perceptrón multicapa (MLP), fue usado para estimar los valores del THI en los niveles de las alturas 1334 y 1338 m en MG y MN, respectivamente. Con base en la temperatura y en la humedad relativa de los niveles examinados a baja altitud (650 m en MG y 676 m en MN), teniendo en cuenta el tiempo de medición real (ATM). Los resultados del desarrollo y aplicación del modelo ampliado MLP indicaron una estimación más precisa de los valores THI en los estudios de las dos regiones durante un periodo de todo el día, comparado con la aplicación MLP sin el uso del ATM. También, el modelo ampliado, examinando el día entero, mostró estimaciones más precisas de los valores THI en el MG comparados con el MN. De manera similar, este modelo proporcionó una mejor estimación por separado del periodo, tanto durante el día (09h00min-20h00min) y durante la noche (21h00min-08h00min) en comparación con las estimaciones respectivas del THI, tomando en cuenta sólo la temperatura del aire y la humedad relativa como parámetros de entrada. Adicionalmente, la ampliación del modelo MLP fue mucho más eficiente para estimar los valores THI durante las horas del día, comparado con las horas de la noche en ambos MG y MN. También el modelo ampliado MLP fue capaz de estimar mejor los valores de THI en la clase Caliente en MG, como así mismo en la clase Confortable en MN.
Genetic algorithms we prove that no matter how close we are to the end of generations to converge to the optimal solutions are needed to reduce the rate of mutation operator. This technique will also lead to changes in the network structure, which is considered one of the most important features of our algorithm. The GANN method with 98.9% accuracy is more better that other introduced methods in this paper. We suggest the machine learning and soft computing methods for future work. Our proposed method can be a model for other problems in data mining. So our approach have high accuracy in prediction stroke disease, also it can be very effective for other data mining problems.
Extensive work has been reported on optimization of process parameters for EDM process across the world. In the present work, an effort has been made to explore the potentialities of application of Regression Analysis and ArtificialNeuralNetwork in EDM process for the case reported. The present paper discusses the investigations carried out on Electro Discharge Machine to find optimum setting of process parameters for a particular tool-work piece combination. Investigation has been focused using four levels of current and three levels of tool diameter to find the optimum values of output quality characteristics viz. Material Removal Rate (MRR) and Surface Roughness. The parameter combination is worked out using Full Factorial Design of Experiment methods (DOE). The experiments carried out using this method take on all the possible combinations of different levels across all the factors and hence allow studying the effect of each factor on the response variable. In present investigation, experiments are conducted for all possible combinations and based on the observed readings MRR and Surface Roughness are measured and regression equations are derived for the two output quality characteristics under study. The present investigation was further extended with the application of ANN. The ANN program is based on Multi layer perceptrons (MLPs) that are layered feed forward networks typically trained with static back propagation. The ANN analysis uses architecture that consists of 2 input and 2 output processing elements and a hidden layer. The network training is carried out and then trained network is tested with few experimental results which are not used during training. Prior to testing, network is also validated with the set of data. The results obtained during the study are critically discussed and reported.
maximum absolute error at each value of L and W (patch dimensions) antenna is estimated for the random values of input parameters but in specified range i.e. the range for which network is trained. Various transfer functions are used for training the network and average minimum MSE on training and CV data is measured. It is observed that tansig as shown in Figure 6 is most suitable transfer function for the present work. The MLP neuralnetwork is trained using learning rules namely Levenberg-Marquardt (LM), Scale Conjugate Gradient Back propagation (SC- GBP), Gradient of Fletcher Reeves algorithm (GFR), Quasi Newton algorithm (QN), Bayesian Regularization Algorithm (BR) and Adaptive Gradient Decent (AGD). Minimum MSE and maximum absolute error is measured on training and test data and is indicated in Table 1 and Figure 7. It is concluded that Levenberg- Marquaradt is most suitable learning rule for our neuralnetwork with 3-20-2 structure. For generalization the randomized data is fed to the network and is trained for different hidden layers. It is observed that MLP with single hidden layer gives better performance as shown in Figure 8. The number of Processing Elements (PEs) in the hidden layer is also varied. The network is trained and minimum MSE is obtained when 20 PEs are used in hidden layer as shown in Figure 9.
One of major challenge facing the healthcare organization today is the provision of quality services at affordable costs. The quality of serviceof service implies diagnosing and administering the patients correctly. A majority of areas of health services such as prediction of heart attack, effectiveness of surgical procedures, medical tests, medication, effectiveness of medical treatments can be estimated by the application of ANN. ArtificialNeuralNetwork (ANN) has extensive application to biomedical systems. Neural networks learn by example, so the details of how to identify diseases are not needed. It requires only set of examples representative of all the variations of the diseases. Of course, the examples are to be selected very carefully if the system needs to identify or predict the disease reliably and efficiently.
Artificialneuralnetwork has been proved to have the characteristics of arbitrary approximation to nonlinear system. Combining artificialneuralnetwork with chaotic system, some scholars have proposed some control methods of chaotic system based on artificialneuralnetwork, and the chaos in chaos system Behavior to implement effective control. Tan Wen et al proposed a chaotic motion based on feed forward back propagation neuralnetwork to control nonlinear systems. Based on the OGY method, the improved BP learning algorithm combined with the BP algorithm of the variable learning rate and the BP algorithm with variable learning rate is used to stabilize the attractor embedded in the unstable chaotic orbit to return to the stable point. Alsing uses a back propagation BP network to stabilize the unstable periodic orbits of embedded chaotic systems. Chen - Teng Lin proposed a GA - based re - excitation learning neuralnetwork controller, which does not need to know the equilibrium point of chaotic system, and does not need the output data of the system to stabilize the system to high - cycle orbit.
Fig 6: Supervised architecture of ANN. Blunder back propagation (BP) neural system was connected for assignment of deficiency in framework. Any way moderate speed training work and furthermore the weaknesses of local optima cause the presentation of extra force issue for disservice goals. Radial basis operates (RBF) neural system envelops a quicker learning pace and also the adaptability of optional work estimation. (6,7) For resolution improper problems, neuralnetwork topologies square measure to be altered and there is a necessity to retrain the network. we tend to square measure able to use beingness capability of multilayer perception (MLP) and generalized regression neuralnetwork (GRNN) for fault estimation in grid.
Neuralnetwork based fraud detection methods are most popular. An interconnected group of artificial neurons is contained in artificialneuralnetwork. The functions of the brain especially associative memory and pattern recognition are responsible to motivate the principle of neuralnetwork. In neuralnetwork similar patterns are identified, future values or events are predicted which are based upon the associative memory of the patterns. Neuralnetwork has widely useful in classification and clustering. It has main advantage over other techniques: the neuralnetwork model is capable of learning from the past and thus, results can be improved as time passes. Also, the rules can be extracted in this model. Moreover the future activity can be predicted on the basis of the present situation. By utilizing neural networks, efficiently, it will become easy for the banks to detect fraudulent use of a card in faster and more efficient way. Amongst the study of credit card fraud that has been reported, it was observed that most have paid attention on using neural networks. Nonlinear statistical data modeling tools are neural networks. Through these networks, complicated input-output relations can be easily depicted. A neuro-fuzzy system uses a learning algorithm that is derived from neuralnetwork theory. This system determine parameters : Fuzzy set and fuzzy rules by processing data samples. This is a fuzzy based model developed in the theory of neural networks using a learning algorithm. The heuristic learning process is based on local data and only creates local changes in the underlying fuzzy structure. This can be regarded as a neural feedback network of three layers. The first layer is input variable, the center layer is hidden and it represents fuzzy rules
Interpolation is one of the most widely used methods of modeling. There are two main types of spatial interpolation methods: deterministic and geostatistical. A deterministic approach, in which results are accurately determined by known relations between states and events, without any possibility of random variation, use methods that calculate unknown values based on the degree of similarity. The methods of geostatistical interpolation (Kriging) use the statistical characteristics of the measured spots together with the spatial autocorrelation between them and take into account the spatial configuration of the sample spots at the forecasting site. The accuracy of the kriging methods depends on the density and size of the sampling sites, since these methods are based on interpolation, which requires some data as input. Therefore, to increase the accuracy of the interpolation methods, a more efficient method is required to obtain high-resolution distribution maps. At the present time machine learning methods are increasingly being used, such as artificialneural networks (ANN) and RF.
training sets. A further aspect that should be mentioned re- gards the complementary trend shown by the error of the training and test set, which describe two opposite curves converging to a unique value (Fig. 9). This confirms that a proper fitting of the neuralnetwork yields good generaliza- tion performances (which exactly means that the expected error on unknown patterns should be very close the one ob- tained during the learning phase). To complete the perfor- mance assessment and test the method under additional re- alistic hypotheses, we evaluated the response of the MLP to e.m signals generated by planar surfaces and lossy media. In the former case, further simulations have been carried out employing the already mentioned horizontal reference plane as upper boundary of the layer. In the latter case, instead, we have generated both rough and planar media with a conduc- tivity s L ranging from 0.01 to 10 S/m.
Predicting the defects in software is one of the significant issues in software engineering that contributes a considerable measure toward sparing time in software generation and maintenance process. Essentially, discovering the optimal models for Software Defect Prediction (SDP) has these days transformed into one of the primary objectives of software architects. Since details and restrictions of software development are growing and negative outcomes like, failure and errors diminish software quality and consumer loyalty, delivering error- free software is exceptionally hard. In this work, a Multi-Layer Perceptron (MLP) neuralnetwork is proposed to classify the software defects. To enhance the performance of the proposed MLP, the parameters of the neuralnetwork is optimized utilizing Genetic Algorithm (GA).
Neural networks exist in a variety of different architectures and have been implemented in numerous financial applications. However, the architecture that is most widely used for the analysis of stock markets is known as the Multi-Layer Perceptron (MLP) neuralnetwork. A generic neuralnetwork is built with at least three layers comprising of an input, hidden and output layer. The structure of the input layer is determined by the number of explanatory variables depicted as nodes in the architecture. The hidden layer represents the capacity of complexity in which the model can support or ‘fit’. Moreover, both the input and hidden layers contain what is known as a bias node. The value attributed to this node is a fixed value and is equal to one. Its purpose is similar to the functionality of which the intercept serves in more traditional regression models. The final and third layer of a standard neuralnetwork, the output layer, is governed by a structure of nodes corresponding to a number of response variables. Furthermore, each of these layers is linked via a node to node interconnecting system enabling a functional network of ‘neurons’.