Minimization of Computational Cost in Tree-Based Methods by a Proper Ordering of Splits

Tree-based methods such as CART (Breiuran et al., 1984), CHAID (commercialized by SPSS) or RECPAM (Ciampi, 1993) have a common typical feature: Analogous calculations are performed many times with one and the same portion of data. An unsophisticated implementation of such methods may result in a low computational effectiveness. Various computational enhancements are dealt with in a series of papers by Mola, Siciliano, Klaschka. and Antoch — see, e.g., Mola & Siciliano (1992, 1997), Siciliano & Mola (1996), Klaschka & Antoch (1997), and Klaschka, Siciliano & Antoch (1998).