In this chapter, we began with a hypothetical dataset and highlighted the problem of overfitting. In case of a breakpoint, also known as knots, the extensions of the linear model in the piecewise linear regression model and the spline regression model were found to be very useful enhancements. The problem of overfitting can also sometimes be overcome by using the ridge regression. The ridge regression solution has been extended for the linear and logistic regression models. Finally, we saw a different approach of model assessment by using the train, validate, and test approach and the cross-validation approach.
In spite of the developments where we have intrinsically non-linear data, it becomes difficult for the models discussed in this chapter to emerge as useful solutions. The past two decades has witnessed a powerful alternative in the so-called Classification and Regression Trees (CART). The next chapter discusses CART in greater depth, and the final chapter considers modern development...