Page 194 - DCAP208_Management Support Systems

Unit 11: Neural Networks

10. The ................... function combines the inputs coming into a neuron from other neurons/ Notes
sources and then produces an output based on the choice of the transfer function.
11. A ................... is a hurdle value for the output of a neuron to trigger the next level of
neurons.
12. The ................... learning algorithm is the standard way of implementing supervised training
of feedforward neural networks.

11.2 Development of Neural Network-based System

Although the development process of ANN is similar to the structured design methodologies of
traditional computer-based information systems, some phases are unique or have some unique
aspects. In the process described here, we assume that the preliminary steps of system
development, such as determining information requirements, conducting a feasibility analysis,
and gaining a champion in top management for the project, have been completed successfully.
Such steps are generic to any information system.

As shown in Figure 11.9, the development process for an ANN application includes nine steps.
In step 1, the data to be used for training and testing the network are collected. Important
considerations are that the particular problem is amenable to neural network solution and that
adequate data exist and can be obtained. In step 2, training data must be identified, and a plan
must be made for testing the performance of the network.
In steps 3 and 4, a network architecture and a learning method are selected. The availability of a
particular development tool or the capabilities of the development personnel may determine
the type of neural network to be constructed. Also, certain problem types have demonstrated
high success rates with certain configurations (e.g., multilayer feedforward neural networks for
bankruptcy prediction). Important considerations are the exact number of neurons and the
number of layers. Some packages use genetic algorithms to select the network design.

There are parameters for tuning the network to the desired learning-performance level. Part of
the process in step 5 is the initialization of the network weights and parameters, followed by the
modification of the parameters as training-performance feedback is received. Often, the initial
values are important in determining the efficiency and length of training. Some methods change
the parameters during training to enhance performance.
Step 6 transforms the application data into the type and format required by the neural network.
This may require writing software to preprocess the data or performing these operations directly
in an ANN package.

!

Caution Data storage and manipulation techniques and processes must be designed for
conveniently and efficiently retraining the neural network, when needed.
The application data representation and ordering often influence the efficiency and possibly the
accuracy of the results.
In steps 7 and 8, training and testing are conducted iteratively by presenting input and desired
or known output data to the network. The network computes the outputs and adjusts the weights
until the computed outputs are within an acceptable tolerance of the known outputs for the
input cases. The desired outputs and their relationships to input data are derived from historical
data (i.e., a portion of the data collected in step 1).
In step 9, a stable set of weights is obtained. Now the network can reproduce the desired outputs,
given inputs such as those in the training set. The network is ready for use as a stand-alone

LOVELY PROFESSIONAL UNIVERSITY 187