Further descriptors and regression methods

Let us check whether even better linear models are obtained by adding geometrical descriptors to our descriptor pool. Upon addition of the 35 geometrical indices from Appendix A.3 and calculation of best linear models containing n = 1. 5 descriptors, a geometrical descriptor enters the model only once  

The algorithm for growing regression trees uses its own logic for selection of predictors (see Subsection 6.2.4). For this example a regression tree of 9 terminal nodes was obtained that used SCg, SC14, SC and had - 0.84386. Due to then- 

linear regression with best subset search seems to be well-suited for modeling BPs of decanes. 

---