Top-down induction of decision trees is the most popular technique for classification in the field of data mining and knowledge discovery. Quinlan developed the basic induction algorithm of decision trees, ID3 (1984), and extended to C4.5 (1993). There is a lot of research work for dealing with a single attribute decision-making node (so-called the first-order decision) of decision trees. Murphy and Pazzani (1991) addressed about multiple-attribute conditions at decision-making nodes. They show that higher order decision-making generates smaller decision trees and better accuracy. However, there always exist NP-complete combinations of multiple-attribute decision-makings.;We develop a new algorithm of second-order decision-tree inductions (SODI) for nominal attributes. The induction rules of first-order decision trees are combined by 'AND' logic only, but those of SODI consist of 'AND', 'OR', and 'OTHERWISE' logics. It generates more accurate results and smaller decision trees than any first-order decision tree inductions.;Quinlan used information gains via VC-dimension (Vapnik-Chevonenkis; Vapnik, 1995) for clustering the experimental values for each numerical attribute. However, many researchers have discovered the weakness of the use of VC-dim analysis. Bennett (1997) sophistically applies support vector machines (SVM) to decision tree induction. We suggest a heuristic algorithm (SVMM; SVM for Multi-category) that combines a TDIDT scheme with SVM. In this thesis it will be also addressed how to solve multiclass classification problems.;Our final goal for this thesis is IDSS (Induction of Decision Trees using SODI and SVMM). We will address how to combine SODI and SVMM for the construction of top-down induction of decision trees in order to minimize the generalized penalty cost.