PaceRegression builds linear regression models using the technique of Pace regression ( Wang and Witten, 2002). It repeatedly applies standard linear regression to subsamples of the data and outputs the solution that has the smallest median-squared error. LeastMedSq is a robust linear regression method that minimizes the median (rather than the mean) of the squares of divergences from the regression line (see Section 7.5, page 333) ( Rousseeuw and Leroy, 1987).

Technically, LinearRegression implements ridge regression, which is described in standard statistics texts. The implementation has two further refinements: a heuristic mechanism for detecting collinear attributes (which can be turned off) and a ridge parameter that stabilizes degenerate cases and can reduce overfitting by penalizing large coefficients. Both methods use a version of the AIC termination criterion discussed in Section 6.7 ( page 266). LinearRegression performs standard least-squares multiple linear regression and can optionally perform attribute selection, either greedily using backward elimination (see Section 7.1, page 311) or by building a full model from all attributes and dropping the terms one by one, in decreasing order of their standardized coefficients, until a stopping criteria is reached (this method was described in a slightly different context in Section 6.6 under Pruning the Tree). Output of SimpleLinearRegression for the CPU performance data.

Weka jar with spegasos