Neural networks solve three major classes of problems:
Prediction: The network estimates a number from a set of input values using pattern recognition that the network learned from the training set of data. Examples of prediction problems include forecasting the next day’s closing price of a stock, interest rate or currency rate, estimating the number of customers for a bank or restaurant, or predicting sales.
Simple Classification: The network estimates an output value (which represents a simple classification) using pattern recognition that the network learned from the training set of data. Examples of simple classification problems include identifying a buy, hold or sell signal, and identifying a bull or bear market. To identify a simple classification, create an indicator that returns a simple numeric classification and allow the neural network to predict this indicator. For the buy, hold, or sell example create an indicator that returns 1,0, or -1 for a buy, hold, or sell. For the bull or bear market example, create an indicator that returns 1 or 0 for a bull or bear market.
Complex Classification: A series of networks each estimate an output value (which represents a category) using pattern recognition that the networks learned from the training set of data. An example of a complex classification problem is identifying whether to invest in the stock, bond, commodity, currency, or option market. To identify a complex classification, create a series of indicators that return a simple numeric value for each category and allow a different neural network to predict each indicator. For the market example, create five indicators that return 1 or 0 for investing or not investing in the respective market.