Regression analysis with interval model by neural networks
Hisao Ishibuchi, Hideo Tanaka · 1991
Proposes a simple method for determining a nonlinear interval model using neural networks from the given data. An interval model whose outputs approximately include all the given data is determined by neural networks. Since an interval model can be represented by two real-valued functions corresponding to its upper and lower limits, the authors propose two learning algorithms of neural networks to determine the two functions. The cost function to be minimized in each algorithm is a weighted sum of squared errors between actual outputs and target outputs. The weight (i.e. penalty) for each squared error is specified at each presentation depending on whether the actual output from the neural network is greater than or less than the corresponding target output.>