Artificial Neural Networks

Introduction

Since the emergence of the computer intelligence, the art and the business of doing business has undergone dramatic changes. Companies continue to adopt more knowledge-based systems. The knowledge-based systems coupled with advanced artificial neuro-computing have become essential components of business intelligence. Neural networks work by imitating the human neurons and works on stimulus from the outside world. Neural networks using artificial intelligence are artificial neural networks. Artificial neural networks involve distributed information-processing systems and powerful general-purpose software tools, composed of numerous simple computational elements interacting across weighted connections (Eric, 2001). Artificial neural networks exhibit distinct features including the ability to learn complex patterns of information and generalize the learned information and are used to make numerous tasks such as prediction, classification, and clustering.

Elements of ANN

In terms of structure, a neural network consists of processing nodes known as neurons that accept values from other neurons through input arcs. These neurons process the input through transfer function and then release the output to other neurons using output arcs. Each of the ANN has a collection of numerous neurons in layers. A typical neuron consists of three layers that include the input, intermediate, and the output. Additionally, they have a hidden layer of neurons that takes input from the previous layer and convert those inputs into outputs for additional processing. In a neural network, the knowledge is stored in the weight associated with each connection between two neurons (Hossein & Khairil, 2011).

An artificial neural network consists of numerous single processors that interact with a dense web of interconnections. The primary functions of computing outputs that are sent to other processing elements or outside the neuron. The processing unit determines its output value by using a transfer function. The second function is to update a local memory. Depending on the connection method and the different information decisions in the network, the neural network can be categorized into two. The first category includes feed forward information transfer that has only forward information transfer but no feedback information. The second category is the feedback neural network that has both forward transfer of information and also reverse transfer (Hossein & Khairil, 2011).

Feed forward neural networks consist of one input layer, hidden layers, and one output layers. The neurons of each layer only accept output information coming from the neurons of the forward layer. Backpropagation is the most common paradigm in business applications of neural networks. A backpropagation-based neural network consists of an input layer, an output layer, and some hidden layers.

Artificial neural networks learning

Neural networks learn to recognize patterns. They achieve learning through repeated minor modifications to select neuron weights. A typical neural network starts with randomized weights for all neurons. As such, they do not know anything in the beginning, and they have to be trained. Once a neural network is trained, and it finds the necessary outputs to a particular input. However, it cannot be guaranteed that a neural network will produce the desired output pattern. Neural networks learn through supervised or an unsupervised learning process.

A supervised learning process has a desired output or a target pattern. While learning different input patterns, the weight values are changed automatically until their values are balanced such that each output will lead to the desired output. The supervised learning involves tow learning algorithms. They include the forward, and back-propagation, learning algorithms.

An unsupervised neural network has no target outputs. During the learning process, the neural cells organize themselves into groups, depending on the input pattern. A single neural cell receive the incoming data or input and also influences other cells in its surrounding. The aim is to group neural cells with similar functions close. The self-organization learning algorithms tend to identify patterns and relationships in the data.

Artificial neural network techniques

There are different types of artificial neural networks. However, the commonest kinds include Perceptron, multi-layer-perceptron, backpropagation net, Hopfield net, and Kohonen feature map. Rosenblatt F (1959) introduced the Perceptron. A simplified version of the Perceptron consists of two neuron layers that accept only binary input and output values. Perceptron learns through supervised learning, and the net can learn basic logical operations. Perceptron is commonly used to classify patterns. Marvin M and Seymour introduced the multi-layered Perceptron in 1969. The multi-layered Perceptron is an extension of Perceptron. It has a hidden neuron layer between the input and output layers. Due to its extended structure, a multi-layered Perceptron can learn every logical operation. In 1982, JJ Hopfield introduced the Hopfield Net. The Hopfield Net consists of a set of neurons, where every neuron connects to each other. The input and output neurons are the same. The core application of Hopfield Net is the storage and recognition of patterns. The Kohonen Net has neurons that compete. Hinton and Williams introduced the backpropagation Net in 1986. Backpropagation is similar to the multi-layer perceptron. However, it uses backpropagation-learning algorithms.

Artificial neural networks applications in business

Neural networks have numerous applications. They can be used to learn to predict future events based on the patterns that they observe in the historical training data. They can learn to classify unseen data into pre-defined groups based on characteristics observed in the training data. Artificial neural networks can learn to cluster the training data into neural groups based on the similarity of characteristics in the training data. In comparison to conventional statistical tools, the core advantage of artificial intelligence is that they can learn to recognize patterns in data sets. They are flexible in dynamics environments, and they can build models when conventional approaches fail. Also, they are less restrictive assumptions on the underlying distribution. Additionally, ANNS have complex systems that can collect massive data.

Artificial neural networks have multiple applications in the business domain. In the accounting sector, ANNS can be used to identify tax frauds and enhancing auditing by finding irregularities. In the finance sector, ANNS apply to the verification of signature and bank notes. They are applicable in mortgage underwriting and foreign exchange rate forecasting. ANNS are also applicable to country risk rating and predicting stock initial public offerings. They apply to bankruptcy prediction and customer credit scoring. Additionally, they are used in credit card approval, fraud detection, and stock and commodity selection. They also have other wider applications in forecasting economic turning points, bond rating, and trading.

ANNs are also applicable to human resource management. It can be used to predict employees’ performance and behavior. ANNS are applicable in determining personnel resource requirements. They are also applicable in the marketing sector to classify consumer spending patterns, new product analysis and identification of customer characteristics (Francisco, Ignacio, Santiago & Francisco, 2010). Artificial neural networks are applicable in forecasting sales and conducting targeted marketing.

Nikhil and Gupta (2010), explored the application of ANNS in bankruptcy prediction, credit card fraud detection, and financial auditing. Bankruptcy prediction is an important knowledge of banks. Banks need the capacity to predict the possibility of default of a potential counterparty before they can offer loans. Banks can use ANNS to make sounder lending decisions thus resulting in significant savings. Banks can use structural or statistical approach. The structural approach relies on modeling the dynamics of interest rates and the characteristics of the firm. The statistical approach use inputs from various financial ratios such as the working capital, retained earnings and market capitalization.

With the expansion of modern technology, bank fraud is increasing causing billions loss. As such, methodologies to detect fraud are essential. Statistics and machine learning offer effective technologies for fraud detections. They are useful in credit card frauds detection. Fraudsters can perpetuate credit card frauds in various ways including simple theft, application fraud, and counterfeit cards.

ANNS are useful in stock market applications. Global financial markets involve trade risks, swap risks and a great amount of uncertainty. As such, the role of accurate prediction is important. Neural networks have found support from portfolio managers, investment banks, and trading firms. As such, most multinationals banks have departments to the implementation of neural networks. In stock prediction, neural networks are useful in stock prediction. For example, Lee (2011), proposed back propagation algorithm using artificial neural networks for stock prediction. Additionally, scholars are identified various models for stock prediction problems. These models use various models including the time series model, recurrent neural network and feed-forward neural network methods (Francisco, Ignacio, Santiago I & Francisco, 2010). These models are used to study the relationships between different technical and economic indices and the decision to buy or sell stocks.

Information technology also affects the nature of the auditing process. ANNS has numerous applications in the auditing process including material errors, management fraud, and decision support. Also, ANNS has applications in control risk assessment, financial distress problems and audit fee (Jovita & Rimantas, 2009). Due to the increased competition and the need for faster and better information and data for decision characterize the current business environment. Material error applications direct auditor’s attention to those financial account values where the actual relationships are not consistent with the expected relationship.

Jovita and Rimantas (2009), explored the use of adaptive business intelligence to improve business rules management. They noted that companies need to respond fast to the changing market, interpret numerous data and be able to make effective decisions. However, these companies have to contend with complex business problems. Some of the major characteristics that make making decision complex include the dynamic business environments, the complexity of the challenge and the large number of possible solutions. Additionally, they have to contend with incomplete information or logical data gaps. These characteristics make the decision-making process complex. Artificial neural networks facilitate managers to make decisions in such dynamic environments.

References

 Eric T N (2001). AI surveying: Artificial intelligence in business. De Montfort University.           

Francisco G., Ignacio S., Santiago I & Francisco  (2010). Use of Artificial Neural Networks to Predict The Business Success or Failure of Start-Up Firms.

Hossein H & Khairil A (2011). Artificial Neural Networks’ Applications in Management. World Applied Sciences Journal. Vol.14 (7); 1008-1019

Jovita N & Rimantas B (2009). Improving Business Rules Management Through The Application Of Adaptive Business Intelligence Technique. Information Technology And Control, 2009, Vol.38, No.1

Nikil B & Gupta M (2010). Application of Artificial Neural Networks in Business Applications.

Sherry Roberts is the author of this paper. A senior editor at MeldaResearch.Com in nursing writing services. If you need a similar paper you can place your order from best custom term papers.