Fuzzy Logic and Neural Networks: Basic Concepts and Applications

Free download. Book file PDF easily for everyone and every device. You can download and read online Fuzzy Logic and Neural Networks: Basic Concepts and Applications file PDF Book only if you are registered here. And also you can download or read online all Book PDF file that related with Fuzzy Logic and Neural Networks: Basic Concepts and Applications book. Happy reading Fuzzy Logic and Neural Networks: Basic Concepts and Applications Bookeveryone. Download file Free Book PDF Fuzzy Logic and Neural Networks: Basic Concepts and Applications at Complete PDF Library. This Book have some digital formats such us :paperbook, ebook, kindle, epub, fb2 and another formats. Here is The CompletePDF Book Library. It's free to register here to get Book file PDF Fuzzy Logic and Neural Networks: Basic Concepts and Applications Pocket Guide.

III — Evaluation computing and other aspects of soft computing. First Break, June Only logged in customers who have purchased this product may leave a review. Publisher EAGE. Description Additional information Reviews 0. Description Soft computing techniques are emerging as valuable methodologies capable of solving a wide spectrum of complex geoscientific problems. Search for books, journals or webpages All Pages Books Journals. View on ScienceDirect. Authors: Erdal Kayacan Mojtaba Khanesar. Paperback ISBN: Imprint: Butterworth-Heinemann. Published Date: 17th September Page Count: For regional delivery times, please check When will I receive my book?

Sorry, this product is currently out of stock. Flexible - Read on multiple operating systems and devices. Easily read eBooks on smart phones, computers, or any eBook readers, including Kindle. When you read an eBook on VitalSource Bookshelf, enjoy such features as: Access online or offline, on mobile or desktop devices Bookmarks, highlights and notes sync across all your devices Smart study tools such as note sharing and subscription, review mode, and Microsoft OneNote integration Search and navigate content across your entire Bookshelf library Interactive notebook and read-aloud functionality Look up additional information online by highlighting a word or phrase.

Institutional Subscription. Free Shipping Free global shipping No minimum order. Powered by. You are connected as. Connect with:. The bias nodes always set as equal to one. A bias major function is to provide node with a constant value that is trainable, in addition to the normal inputs received by the network node. Importantly, a bias value enables one to move the activation function either to the right or the left, that can be analytical for ANN training success. When the NN used as a classifier, the input and the output nodes will match input features and output classes. However, when the NN is used as a function approximation, it generally has an input and an output node.

However, the number of designed hidden nodes essential greater than those of input nodes. Given this description of neural networks NNs , how its work, and their real-world applications and uses, indeed, NNs have wide applied to real-world problem in business, education, economics and in many aspects of life problems. NNs are also applicable to optimization method [21] intrusion detection [ 22 , 23 ] and data classification [ 24 , 25 , 26 ].

Classification regarded as a form of difficult optimization challenge. Most researchers applied machine learning ML techniques in solving classification problem [ 27 , 28 ]. NNs are excellent identifier of trends in data and patterns [29] , they are suited for forecasting and prediction needs including those items listing with references to Supplementary Table 1.

It summarizes neural networks application in practice which integrated into many areas like modeling, classification, pattern recognition, and prediction. Prediction of financial stability is useful in economic, management and development of any nation, which is beneficiary for analyzing the monetary value of any economy. Moreover, ANNs have been used successfully in the prediction of banks success or failure and stock market estimation.

Likewise, it is used extensively in forecasting of weather and climatic change which is helpful in human safety and security of properties such as buildings, environment, installation, houses, and transportation. Furthermore, ANNs applied successfully to different areas of agriculture like remote sensing, particularly in the crop type classification and crop production estimation. Therefore, Supplementary Table 1 provides a general review of the wide scope of problems that this neuro-Intelligence system can currently address.

ANN can be classified as depicted in Fig. A feedforward neural network FFNN is a machine learning classification algorithm that made up of organized in layers that are similar to human neuron processing units. In FFNN each unit in a layer relates to all the other units in the layers. These layers connections with units are not all equal because each connection can have a different weight or strength. The weights of the network connections measure the potential amount of the knowledge of the network.

Also, NN units are known as nodes. The information processing in the network involves data entry from the input units and passes through the network, flowing from one layer to another layer until it gets to the output units. When NN operate normally, that is when its acted as a classifier, then there will be no feedback between layers [30].


  1. Finding Home;
  2. Officers Guide to Fire Service E.M.S.
  3. Pro ASP.NET 3.5 in C Sharp 2008.

In FFNN, information transmitted only in one direction, that is from the input nodes, to the hidden nodes, if any, and then to output nodes. With this behaviour, they are called feedforward neural networks [31].

An Introduction to Fuzzy Logic

Examples of FFNNs is single layer perception and multilayer perceptron. Example of a two-layered network is 3 input units, 4 units with a hidden layer and 5 units of output layer as circles respectively in Fig. A hidden layer is neither an input or output layers, that is Fig. It is clear that a layer only connects to the previous layer. FFNN applications is classified into two such as control of dynamical systems [ 32 , 33 ], and spaces where the classic machine learning techniques are applied [34].

NNs with two or more hidden layers are called deep networks because the network has become complex with more than 1 hidden layer. That means FFNN can logically handle task according to first come first serve bases of inputs. Feed-backward NN can applied to tasks like un-segmentation, and pattern recognition connected handwriting recognition.

Fuzzy Logic and Neural Networks by Chennakesava R. Alavala | Waterstones

Feed-backward neural network application areas include mathematical proofs, seismic data fitting, medicine, science, engineering, classification, function estimation, and time-series prediction, etc. In feedback NNs or backpropagation, connections between nodes produced a coordinated graph in sequence. The coordinated graph in sequence allows feedback NNs to demonstrate dynamic terrestrial behaviour for a time sequence. Examples are Kohonen's self organizing map and recurrent neural network RNN. RNN referred to a standard kind of neural network which extended over time, with edges that feed into the next time step rather than feeding into the next layer concurrent time of step.

RNN is constructed to sequences recognition, for instance, a text or a speech signal. It has cycles within that indicates presence of short-memory in the net. Unlike a recurrent neural network, an RNN is like a hierarchical network where the input need processing hierarchically in the form of a tree because there is no time to the input sequence.

Artificial intelligence AI has existed over many decades, and the field is wide. The distinction between deep learning and neural networks like feedforward NNs and feed backward NNs lies in their characteristic. Deep learning has more complex ways of connecting layers, also has more neurons count than previous networks to express complex models, more also with more computing power to train and further has automatic extraction of the feature. Therefore, DL defined as a NN with a broad variables and layers with a single basic network architecture of unsupervised pre-trained networks, convolutional NNs, recursive NNs, and recurrent NNs.

The technological development in the field of AI has expanded over time as showed in Fig. DL methods have been found to be fitting for big data study with remarkable success in its applied to speech recognition, computer vision, pattern recognition, recommendation systems, and natural language processing [36]. Nowadays, the innovation of DL in image identification, object detection, image classification, and face identification tasks have great success. The review provided a comprehensive knowledge on four DL architectures like, deep belief network, restricted Boltzmann machine, autoencoder, and convolutional neural network CNN.

CNN is designed to recognize images by having convolutions within, that can recognize the image of an object. CNN has multiple layers; including fully-connected layer, pooling layer, convolutional and non-linearity layers. The fully connected layers and convolutional layers have parameters, however non-linearity layers and pooling do not have parameters.

Particularly, in the applications to image data, like the most extensive image classification dataset, natural language processing, and computer vision. The major concept of deep learning DL is learning data representations by increasing the quality of handling the ideas rather than events abstraction levels. Mostly in all levels, a significant amount of quality ideas or abstraction representation at a advance level are known through definition regarding fewer quality ideas or non-representations at the basic levels.

This type of stages of learning, growth or hierarchical process of learning is superb because it can enable a system to fathom complex or multi-complex presentations accurately from raw data [ 38 , 39 ]. This superb characteristic is making deep learning applicable to different fields. To fast tract classification, recognition of patterns in biological data, many methods of artificial intelligence particularly the machine learning has been proposed [ 38 , 40 , 41 ]. Machine learning has become the latest model to digital evolution, that is making computing processes more cost-effective, efficient, strong, dependable and reliable.

Conventional machine learning techniques widely classified into two sets that is, supervised and unsupervised. The supervised learning has the capability of classifying objects in a pool with possibility of given features or attributes nor annotations. A typical example of supervised learning is when students have written an exam, and having the exam marked by the teacher and shown which questions the students answered incorrectly.

After being shown the correct answers, the students are expected to learn then understand how to answer those questions correctly. But the unsupervised learning methods form clusters or groups between and among the objects in an area to identify likeness, then use similarity for classifying unknowns.

Example of unsupervised learning is the man that is learning how to ride a car by himself. He will start by entering the car and start the engine with the ignition key, then put down the clutch and put the car in first gear and press the accelerator for the car to move forward and then manage to control the steering. He continually practices the driving steps in a large open field and gradually master the driving technique, and then over time start to enter the road to drive skillfully.

The reinforcement learning RL category, enables a system or an agent learn from the previous experiences gains in the environment through interaction and observing the results of these interactions. The interaction helps to mimic or imitates the basic pattern in which humans and animals learn. An agent of RL can act, and each act influences the agent's future condition, a scalar reward signal measures RL success.

RL goal is to choose actions that maximize future gain. In nutshell, DL, is a general-purpose framework in making decision. A framework that present learning given require objective from raw inputs by applying minimal domain knowledge. For instance, RL completely bypasses the problem of rules learning in a task. The learning agent learns by participating in the given activity. An example is a game of Chessboard, to determine the best action play move, the players need to think about various possibilities and strategies.

The amount of possibilities is potentially big that it is not possible to perform a brute-force search. However, if a machine is to be built to play such a game using traditional techniques, there will be a necessity to specify many rules to cover all these possibilities. Reinforcement learning completely bypasses this problem. Meanwhile, some common unsupervised methods include k-means [52] , expectation maximization [53] , and autoencoders [54].

Other unsupervised approaches in literature are density-based [55] , self-organizing maps [56] , clustering [57] and fuzzy [58]. Kernel-based neural networks methods are a class of algorithms for pattern analysis, like SVM support vector machine. Kernel functions or algorithms used in shallow architectures, for example support vector machines SVMs , or multilayer kernel machines MKMs.

The study highlighted the usefulness of kernel-based neural networks in the applications to optical character recognition and DNA analysis.


  1. The Chinese Sisters of the Precious Blood and the Evolution of the Catholic Church.
  2. 1st Edition.
  3. Description:.

The research by Camps-Valls, and Bruzzone [60] , proposed kernel-based techniques for classification of hyperspectral image, with main characteristics of various kernel-based methods. ANNs has significant advantages over statistical models when both are relatively compared. In ANN models there are no assumptions about data properties or data distribution. Therefore, ANNs are more useful in practical application. Also, unlike some statistical models that require certain hypothesis for testing, ANN models do not require any hypothesis. ANNs are very flexible, data reduction models, encompassing nonlinear regression models, and discriminant models.

More also, unlike the support vector machine, extreme learning machine, and random forest, ANNs are more fault tolerant. That is, they can handle incomplete data and noise, they can solve non-linear problems, Also, trained ANNs, can generalize at high speed and make predictions. Furthermore, ANNs are scalable when relatively compared to the support vector machine, extreme learning machine, and random forest.

Interestingly, DL techniques attracted more research attention since year DL has the built-in ability to solve the defect of traditional paradigms dependent on hand-built materials. Also, DL approaches have been useful in big data technique with performance applications to pattern recognition, natural language processing, recommendation systems, speech recognition, and computer vision. Generally, modeling is the process of presenting a real-world phenomenon or object as a set of mathematical expression. However, neural network modeling is a process representing the way the nervous system functions.

That is, a NN is a simplified representation of how the man brain processes information. Its function by simulating many interconnected processing units that resemble idea versions of neurons. It is the most common optimization paradigm used in optimizing a neural network. Recently gradient descent used in updating weights in a NN model, that is updating and changing the model's parameters in a direction to minimize the Loss function. A study by Ludermir, Yamazaki, and Zanchettin [61] , proposed a new method for NN global optimization. The method combines backpropagation algorithm, annealing, and tabu search, to produce an automation for generating networks with low complexity and high classification.

The results showed that the new method is better than the previous studies obtained by the most used optimization approaches.

Neural Networks and Other Soft Computing Techniques with Applications in the Oil Industry

Recently, researchers synthesized artificial enzymes that functions or operates in the metabolism of living cells. These artificial enzymes used cell's own energy that enable hydrogen gas production from solar energy [62]. ANNs have applied in many ways like in system control [63] , pattern recognition [64] , power systems [65] , robotics control [66] , forecasting [67] , manufacturing [68] , social sciences [69] , Art [70] , optimization [71] , psychological sciences [72] , signal processing [73] etc.

In an optimizer modeling approach to a solution of reusability problem. An example of a practical approach to modeling in the optimizer for a software architecture. It then highlights, a generator analytical expressions of the framework system , an optimization service, the component standard and the pattern of using the designed framework. The result demonstrates how the framework system for the software components can be used in building new generations that optimized environment.

That allow capitalization and reutilization by the combined software packages and optimization algorithms. Thus, the approach showed the possibility of building a global software architecture that optimized systems and components. In measuring a good prediction model to predict the expected outcome, a loss function required. A most popular used method of finding the minimal point of a function referred to as gradient descent.

DNNs are currently among the most popularly used classifiers. In modeling on loss function, a paper by Janocha, and Czarnecki [75] , investigated the effect of loss functions choices on deep models and its learning dynamics, and the resulting classifiers robustness to different effects. Two experiments L1 and L2 performed on classical datasets.

The result shows that L1 and L2 losses are, quite interesting, that justified classification objectives for deep neural nets because it gives probabilistic interpretation regarding expected misclassification. In finding the optimum design that provides a lightweight and high quality at the same time, it is significant to have effective and efficient prediction methods at the initial design stage.

In a research work on weight optimization method, the paper by Yu, and Chi [76] , proposed a weight model optimization in credit evaluation, based on the concept that the optimal is the weight. That is, after empowerment the result of credit evaluation should has maximum discriminating power to differentiate non-default from default customers. The empirical results demonstrate that the discriminating power of credit evaluation was the strongest compared with the three types of weight models, like t-value, mean square error, and variation coefficient. In the recent times various successful used of ANNs emerged in catalysis, meteorology, biology, chemistry, physics, nuclear physics, high-energy physics, and other areas of science.

Nowadays, ANN has found uses in a new area such as in catalyzing especially in the chemical industrial sector. Catalysis is term as the significant energy in the modernization process of chemical industries. It ensures effective, efficient and successful use of finite natural resources, it prevent waste and air pollution, and provides safety for the industrial sector.

Catalysis become the foundation of large scale operations regarding size in chemistry and petrochemistry environment. However, as demand changes, new environmental challenges now require new catalytic solutions. For example, changed in the energy economy has driven an increasing demand for coal and gas, hence given room for new challenges for catalytic technology in the areas like liquefaction in material science [ 77 , 78 ].

Fuzzy Logic and Neural Networks Basic Concepts & Application

Recently, there have been reported cases of ANNs applications to catalysis research in the literature. The reviewed paper by Li, Zhang, and Liu [79] , show how ANNs applied to catalysis helps people in addressing the complex problems and then accelerating the progress of the catalysis utilization. The reviewed paper further showed how ANNs applied in many ways to catalysis prediction, new catalysts design, and understanding of catalytic structures which produced effective result.

In another development research on an accurate description of chemical processes using computational methods like density functional theory DFT. Behler and Parrinello [81] , introduces a new breed of NN model of DFT, that gives the energy as a function of all atomic positions in systems with arbitrary size and in various orders of magnitude that is faster than DFT. This NN approach is generic which can be apply to all kind of periodic and nonperiodic systems.

Recently, a generalized ML input representation was applied by quantification and concentrations of blended solutions in addressing the problem of determining intrinsic trends in CO 2 solubility under a specific condition. The research orchestrated by Li, and Zhang [82] , applied general regression NN GRNN algorithm in fitting intrinsic trends or movement of CO 2 solubility with a minimal amount of experimental data. Which resulted in the average RMSE root mean square error less than 0.

The study has shown that applied generalized input representation, could provide a better comprehension of the inherent trends of CO 2 solubility in a blended amount of solutions. ANNs are useful and applicable in system modeling like in implementing system identification and complex mappings. For instance, the application of NNs to renewable energy challenges have shown tremendous success. In , an investigation by Kalogirou [83] , applied ANN to energy like solar regarding modeling and design of a solar steam generating plant.

The experimental result demonstrates prediction for speed, load, and error. Wavelet networks are been used extensively and effectively in different engineering fields for classification, identification and control problems.

Description

Wavelet networks are feed forward networks that uses wavelets as activation functions. ANNs application to solar radiation data forecasting with adaptive wavelet network has been useful. The most recent work by Li, and Liu [84] , uses an adaptive wavelet network architecture in discovering an appropriate model for forecasting the daily total amount of solar radiation. The daily total amount of solar radiation is considered the most significant in the prediction of the performance of renewable energy like solar, importantly in sizing photovoltaic PV phenomenon of power systems.

That is in the conversion of light into electricity. Also, Li, and Liu proposed a model on the optimization of the solar water heater and performance prediction using a knowledge-based machine learning technique. The result demonstrates that the model generalization that can applied in different locations even for weather data, like ambient temperature and sunshine period. In recent years ANN application to chemistry and physics problems has increasingly popular and successes [85].

Many applications of approximation techniques and standard approaches to data fitting are performing better in NN. NN provide more accuracy with a lower number of adjustable parameters than any other methods. Learning in NNs is understood when it rebuilds hypersurfaces together with a sample points, generalization, and interpolation. NNs apply sigmoidal functions to re-buildings or transformations, stated in most physics and chemistry problems.

Thus, resolving an arbitrary data fitting issue by applying a single layer net architecture if there is no restriction in sigmoidal functions applied [85]. Neural networks application in physics has witnessed a remarkable success. ANN has been found useful in predicting the concentrations of radioactivity. An understanding the levels of radioactivity values with values of other variables in the environment can be apply to train a network to estimate the next levels of radioactivity. Hence, accuracy of the NN method can be better than other approaches in certain monitoring areas.

Nuclear theory main goal is to predict nuclear structure and nuclear reactions from the fundamental theory of high cohesion interactions, and quantum chromodynamics QCD. However, to accurately calculate the properties of atomic nuclei, there are a plethora of theoretical and computational challenges involves. The result demonstrates that FFANN could predict properties of the 6 Li nucleus like the gs energy and the gs point proton rms radius. The result satisfied the ideal physics condition.

An important advantage of the ANN method is that it does not require mathematical relations of the input data and output data. Importantly, ANN applications to Physics and Chemistry problems should be compared with other methods like statistical techniques and data fittings procedures for the performance measure. Many artificial neural network techniques have been adopted in the academia and industries to address the challenges in computer vision, speech and pattern recognitions, face alignment, and detection. These include;. The application of ANNs has become divergence and understood in the capability of its successes in speech or communication recognition.

In the past decades, ML algorithms have applied widely in areas like acoustic modeling and ASR automatic speech recognition [88]. Computer vision aims at making computers to accurately understand and process visual data efficiently like videos [88] and images [ 89 , 90 , 91 ]. Main goal of computer vision is to provide computers with the kind of ability of man brain functionality. Theoretically, computer vision alludes to the logical control which studies how to separate data from images in artificial frameworks.

Sub domains of computer vision include object detection and object recognition, object estimation, object position, event detection, scene reconstruction, image restoration, image editing, video enhancement, and statistical learning. Hence, in computer vision, ANN models are very useful. The recent improvement in deep learning models has given novel ways to deal with the issue in recognition of a pattern or pattern recognition PR. PR is a scientific area that focus in identification of sequence in each input [ 92 , 93 ].

PR is a general concept that surrounds various subdomains such as speech tagging, regression, sequence labeling and classification.

This item is not reservable because:

There are rapidly increasing needs for information processing and output, due to industrial development, that has new trend and challenges to PR. Face alignment plays a role that is significant in diverse visual applications. In recent times ANNs has claimed successes in face alignment [ 94 , 95 , 96 ] and face recognition [ 97 , 98 , 99 ] and other models [] have shown successes.

Interestingly DL techniques can be applying to explain genetic variants to identify pathogenic variants [ , ]. Usually, combined annotation dependent depletion algorithm is popularly applied to interpret the coding and non-coding variants. Detection in medical diagnosis, security, image objects, financial irregularity, a fault in a system, are being enhanced through ANNs application. Thus, ANN plays an essential role in the detection, particularly when applied to breast cancer [ , , , ].

The performance of ANN can be relatively compared with other approaches in crime detection such as DNA and activity profiling [] and the use of big data for financial crime detection []. Despite the many publications in the utilization of NN in different medical challenges, but there are few reviews study available that explain the architecture in improving the detection methods regarding performance, accuracy, sensitivity, and specificity. Thus, detection capability is commonly known subdomain or computing in computer vision which seeks to understand, locate, classify or differentiate the targeted image objects.

An example during detection tasks, an image can be scanned to know certain special features or characteristics. For instance, using of image detection in medical diagnosis, especially abnormal cells or tissues in medical images. Normally, traditional methods are base on hand-designed features and contrast inference mechanisms. The DL techniques require raw image data only [ , , , ].

Also, DL techniques applied to Glaucoma detection with promising results [ , , ]. More also, ANN has employed in image change and computer vision detection in both civil and military challenges. Recently, image detection applying in remote sensing, disaster evaluation, videoing and surveillance. Furthermore, ANNs or deep learning techniques has been applying to human-robot interaction systems that yielded results [ , , ]. This section highlights discusses, compares, summarizes and critiques more than eighty research articles on artificial neural network model's application to the diverse area of the economy.

The comprehensive review has shown that interest in ANN applications exploded over the past two decades. The results indicate that the selected research articles are recent because they were majorly published between the year — and focuses on developmental and technological issues regarding ANN.

The survey covers various areas of ANN applications such as computer security, network security, science and engineering, medical science, biology, ecology, nuclear industry, electricity generation, management, mineral exploration. Other areas of application are; crude oil fractions quality prediction, crops, water treatment, policy and businesses like banking, insurance, the stock market, money laundering and other financial institutions for crime detection. The field of application varies from science, engineering, socio science, humanity and to art-related fields. From the survey of various articles, it became clear that the artificial neural network application has no boundary.

ANNs has attracted the most attention from researchers in recent time, for instance, the study found that forecasting of crops and animals yield are helpful in agricultural development. Interestingly, this application of ANN has led to an increase in agricultural production which has enhanced food security in many nations. In the past two decades, ANNs have applied in different fields of agricultural, particularly to crop area estimation and classification. Also, agricultural data can be predicted using a neural network. There are diverse approaches proposed for data analysis such as neural network models; examples are; feedforward artificial neural network, back propagation neural networks, probabilistic neural network.

Others include supervised associating networks, multi-layer perceptron neural network architectures, learning vector quantization, and multi-layer neural network. Although, the study by Kitchens, and Harris [] , on ANN application to detection of fraud in insurance business and finance, demonstrates fraud detection process efficiently. However, the result needs comparison with other approaches like data mining, regression model and other statistical models for evidence of successes.

Though, the research by Fanning and Cogger [] , on the application of ANN for fraud detection in management show evidence of success in fraudulent detection using ANN in financial statements. Nevertheless, the result needs to be compared with other current techniques if it performed better or outperformed state-of-the-art detection techniques. More also, even though ANN application is quite novel, one must not ignore that nowadays there are other novels criminal detection techniques with graphical and programming applications like graph-theoretic anomaly detection and inductive logic programming.

Although, ANN tools make it simple, easy and faster for data analysis to enable not only the discovery of a new method for businesses, industrial, and educational topologies but also cross-product, cross-channel and cross-customer performance. Nevertheless, ANN is not necessarily a panacea; there has to be information and communication technology ICT involvement in making the data sources available and, of course, ANN tools do not exclude the need for data cleansing.

But it gives new opportunities to an organization looking to tackle management challenges such as improvement of materials, product, services, financial crime, and so on via new means. With emerging technologies such as data mining, genetic algorithms, hybrid models, mathematical models, big data application especially in crime detection and prevention [ , ]. The evolution of computer and the internet has the influence of rapid technology and digital media growth on every aspect of human lives [ , ].

In addition to extensive interactive software application [].

Researchers can follow adequate artificial intelligence application mechanisms to achieve huge success in the diverse field of endeavours. In nowadays of global computing, there are a plethora of benefits in NNs. Since capability to learn during and after training makes ANN very powerful and flexible. Also, ANN do not require paradigm before performing a specific task, that is, without the need in understanding the external or internal mechanisms of the task before implementation [].

The result of ANNs application to different areas of lives and disciplines as found in the literature is presents in Table 1. Summarized result on ANNs application regarding prediction, pattern recognition and Classification. Table 1 indicates that ANN models are useful in classification, pattern recognition, claustering, optimisation and prediction. The relationships among many areas of ANN applications further presented in Fig. The correlation among the distinct fields further reveals that ANN can apply to any areas of studies, industries, and profession.

The histogram reveals the areas of application of ANN in security, science, engineering, medical science, agriculture, finance, banking, weather and climate, education, environmental, energy, mining, insurance, marketing etc. Therefore, interested researchers can explore the ANN application in these areas or many other emerging areas for future research for better solution to problems in their fields.

Since there is always an algorithm, model, scheme, and framework for any problem. The survey was comprehensive with a discussion on how NN could applied to address human needs. The ANNs application areas considered in the survey include; computer security, medical science, business, finance, bank, insurance, the stock market, electricity generation, management, nuclear industry, mineral exploration, mining, crude oil fractions quality prediction, crops yield prediction, water treatment, and policy.

It is interesting to know that neural network data analysis adds accuracy, processing speed, fault tolerance, latency, performance, volume, and scalability. Many new and enhanced data management and data analysis approaches help in the management of ANN. Creating analytics from the available data that aid in largely prioritizing information and provide its human business value.

The ANN analytics in turn help in combating challenges and mitigate any possible risks. Therefore, based on data analysis factors such as accuracy, processing speed, latency, performance, fault tolerance, volume, and scalability, an evaluation was made of the ANN techniques. Then, proposes that neural-networks models such as FFBP and hybrid model using neural networks are performing better for implementation of human problems when compared to other approaches currently in practice.

Also, the study proposes hybrid neural networks models and genetic algorithms GA for a better performance regarding effectiveness and efficiency. ANN are new computational model with rapid and large uses for handling various complex real world issues. ANNs popularity lies in information processing characteristics to learning power, high parallelism, fault tolerance, nonlinearity, noise tolerance, and capabilities of generalization.

Based on the reviewed literature, some areas of improvement can be suggested to professionals, researchers and newcomer researchers for further research and future research development. Therefore, the following suggestions are stated. Transformation of data. The past and current studies indicate that it is unnecessary to transform or change data not normally or usually distributed that reveal non-regular periodic development or variant. Meanwhile, the modification of heteroscedasticity and trends in data are encourage. Also, data normalization and scale to commensurates with function transfer in output layer.

Determining of appropriate inputs model. Input variables determined with the support of a priori knowledge, using a stepwise model-building method or analytical method like cross-correlation technique. Network geometry choice. One hidden layer may be enough as adequate in most network practical uses. However, mathematical expression could determined the upper bound of hidden layer nodes required to approximate continuous function. Except if nonconvergent techniques, like cross-validation, the connection between the quantity of training and the quantity of hidden layer units likewise should be considered.

The relationship can be investigated with the guide of the rules given in some literature. Researchers can focus on network characteristic at variable specification phase. Also, it is useful to conduct trials in determining required local minimum in the error surface, and oscillations in the R. Forecasting with continuous training, with different step sizes taken in weight space. These steps can be useful when selecting appropriate network parameters like transfer function, momentum, epoch size, learning rate, and error and how many training samples in the network for a case study.

Validation of model. The validation of model is necessary for standardization and practical scenarios of ANNs for optimization of performance. This research has many areas in need of further investigation. Further research is requiring in the following areas:. It would be a good idea for further research to be carried out in the following areas:. Meanwhile, the research on ANNs applications will attain more significant progress in the nearest future. All authors listed have significantly contributed to the development and the writing of this article.