ABSTRACT
This chapter describes the application of Particle Swarm Optimization (PSO) to concrete parametric and non-parametric regression problems. It presents some common issues of practical importance in these problems are identified and possible ways to handle them. The chapter shows how a Swarm Intelligence method such as PSO can help open up new possibilities in statistical analysis. The minimal performance condition used for tuning PSO in parametric regression – where the true signal parameters are known – is not a practical one to use for tuning in non-parametric regression. Statistical regression involves the optimization of a fitness function over a set of regression models. In the general case, analytical minimizations should be carried out first before using a numerical minimization method if a regression model admits this possibility. As the number of breakpoints increases in the regression spline method, so does its tendency to cluster them together in order to fit outliers in the data rather than fit a smooth signal.
