Comparison of piecewise regression and polynomial regression analyses

In this study, it was aimed to compare quadratic and cubic piecewise regression analyses and univariate polynomial regression analysis using both simulated data and real data sets. In the application step of the study, algorithms are created by using R software for simulation practice. Estimation performances of the polynomial and piecewise regression models created using both data sets generated by means of these algorithms and real data sets are compared according to coefficient of determination (R2), mean square error (MSE), Akaike information criteria (AIC) and Bayes information criteria (BIC). At the end of the simulation applications and the real data sets, the R2 values of all piecewise regression models formed with respect to the most suitable knots are higher than those of polynomial regression; MSE, AIC and BIC values were found to be lower. As a result, for all data sets to examine the relationship between dependent and independent variables, the piecewise regression models that are formed according to the knots determined optimally by the number and the position make higher performance estimations than the polynomial regression models. Keywords: Polynomial Regression, Piecewise Regression, Knot, Simulation.