The use of a Gaussian cost function in piecewise linear modelling

Traditionally a least squared error (LSE) function is used when developing a regression model. Recently fuzzy approaches to modelling have become popular with researchers using fuzzy inputs and outputs in their models. Despite the advances in fuzzy modelling, the use of a fuzzy cost function is limited. This paper evaluates a Gaussian fuzzy fit function as a cost function when performing segmented linear regression. The Gaussian function is found to give the modeller greater flexibility in building the regression model and more control over the modelling process. As well, with additional analysis tools, this fuzzy approach allows the optimum number of segments and the knot placement to be determined automatically. Determining these parameters is difficult when performing multidimensional segmented regression using traditional approaches. Although not a panacea, in a broad range of modelling situations, the fuzzy cost function provides superior performance