1. Artificial Neural Network
An Artificial Neural Network (ANN) is a computational model inspired by networks of biological neurons, wherein the neurons compute output values from inputs. All signals can be assigned binary values as either 1 or ā1. The neuron calculates a weighted sum of inputs and compares it to a threshold of 0. If the calculated sum is higher than the threshold, the output is set to 1 or to ā1. The power of the neuron results from its collective behavior in a network where all neurons are interconnected. The network starts evolving: neurons continuously evaluate their output by looking at their inputs, calculating the weighted sum, and then comparing to a threshold to decide if they should fire. This is a highly complex parallel process whose features cannot be reduced to phenomena taking place with individual neurons. One observation is that the evolution of an ANN causes it to eventually reach a state where all neurons continue working, but no further changes in their state occur. A network may have more than one stable state, and it is determined by the choice of synaptic weights and thresholds for the neurons.
ANN is a computational model that is based on a machine learning technique. It works like a human brain neuron system. This machine learning technique follows the same pattern of learning, that is, learning from its past experience and mistakes like mammalian neurons to achieve the target value. An algorithm is designed on the basis of a neural network system to implement a parallel computational power of neurons. ANN learns from its past experience and errors in a nonlinear parallel processing manner using a popular algorithm named āfeed forward and backpropagation.ā The term āfeed forwardā describes how the neural network processes and recalls patterns. In a feed forward neural network, neurons are only connected forward. Each layer of the neural network contains connections to the next layer, but there are no connections back. The term ābackpropagationā describes how this type of neural network is trained. Backpropagation is a form of supervised training. When using a supervised training method, the network must be provided with both sample inputs and anticipated outputs. The anticipated outputs are compared against the actual outputs for given input. Using the anticipated outputs, the backpropagation training algorithm takes a calculated error and adjusts the weights of the various layers backward from the output layer to the input layer to reduce the value of error. The information is delivered to output if it achieves the target; otherwise, it is backpropagated. Hence the name of the algorithm is feed forward backpropagation. The target value will only be achieved if the weighted sum will meet the minimum threshold and hence feed forward or backpropagate for further processing. ANN could be an excellent choice to process large biological data for a more accurate prognosis. The prognostic tools can be designed based on ANN's powerful learning and processing characteristics, which can work perfectly even in a highly probabilistic and noisy environment. The power of the neuron results from its collective behavior in a network where all neurons are interconnected. The network starts evolving; neurons continuously evaluate their output by looking at their inputs, calculating the weighted sum, and then comparing to a threshold to decide if they should fire.
The neuron is the basic calculating entity in ANN processing, which accepts information from a number of inputs and delivers one output by comparing with a threshold value. The computational processing is accomplished by internal structural arrangements that consist of hidden layers and algorithms to deliver a specified output. The learning is based on reinforcement (supervised) and unsupervised (no target) types. The unsupervised mimics the biological neuron pattern of learning.
Basically, ANN is a mathematical model that is used to implement the designed algorithm-based machine learning techniques. ANN communication is performed by calculating the weights of neural inputs, which works on the basis of mathematical operations such as multiplication and addition. Each input received at a nodal point is multiplied with its weights and summed together before activation (firing). In the case of a biological neuron, the information is received at dendrites, processed at soma (cell body), and delivered to axon (output). Similarly, in ANN, the artificial neuron is the basic unit of information reception where the inputs are received and multiplied, summed, and processed via a transfer function before being delivered to the output. An ANN model is so simple and natural that it can handle very complex real-life problems in a nonparallel and distributive way like a biological neural network. The mathematical description of ANN can be understood by the following equation:
where
Xi(t) is the input value at time t,
Wi(t) is the weight of neural input at time t,
F is a transfer function,
Y(t) is the output value at time t.
Note that the transfer function F is to be selected on the basis of the nature of the problem. It mathematically defines the properties of neurons. It can be any step function or nonlinear sigmoid function, depending on the problem. The step function is used to handle classification problems like classifying the benign and malignant state of breast tumors. Similar to a human neuron network, ANN should be trained before it is actually applied to a specific problem. This learning can be supervised or unsupervised in nature.
2. ANN as a Classifier
ANN can be used to classify the complex and noisy biological data for prognosis. For example, in breast cancer tumor data, an ANN classifier can be trained to classify the benign and malignancy states based on descriptors like cell uniformity, clump thickness, size, shape, intensity, mitosis, etc. Its performance is then judged through mean square error and confusion matrices. Data are loaded in terms of feature elements.
The data available for processing under ANN are distributed in categories such as training, validation, and testing. Training data are the actual data offered to a network during training and adjusted according to errors and mistakes. Validation data are used to test the network performance directly and stop the processing in case of overfitting. Out of sample testing is an independent operation and has no effect on ANN operation durin...